Fix: consolidate and extend unicode NFC normalization for filenames and matching

Consolidates the scattered unicodedata.normalize("NFC", ...) calls introduced
by my earlier filename/path normalization work into a single
documents.utils.normalize_unicode() helper, and closes several gaps where
NFD-normalized filenames could still slip through unnormalized:

- Document.get_public_filename() now normalizes, fixing exported filenames
  built from an NFD title/correspondent name (the default, non-format export
  path was not covered by the earlier fix).
- DocumentViewSet.update_version() now normalizes the uploaded filename,
  matching PostDocumentView's existing behavior.
- ConsumerPlugin normalizes self.filename once at consumption time, covering
  the title fallback and Document.original_filename for every document
  source (consume folder, mail, API, barcode splits).
- Workflow trigger and mail rule filename/path matching (documents/matching.py,
  paperless_mail/mail.py) now normalize the document/attachment side before
  comparing, so an NFD filename matches an NFC-typed filter pattern instead of
  silently failing to match.
- WorkflowTriggerSerializer and MailRuleSerializer normalize filter_filename/
  filter_path and the attachment include/exclude patterns once at write time,
  so the read side isn't re-normalizing an already-canonical value on every
  match.

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/run

@@ -1,58 +0,0 @@
-#!/command/with-contenv /usr/bin/bash
-# shellcheck shell=bash
-declare -r log_prefix="[init-compile-bytecode]"
-
-# PYTHONDONTWRITEBYTECODE=1 is set for the whole container. This unit compiles a
-# scoped set of libraries anyway, to speed up startup without bloating image size.
-
-# Handle the people using a read only file system
-if [[ "${S6_READ_ONLY_ROOT}" == "1" ]]; then
-	echo "${log_prefix} S6_READ_ONLY_ROOT=1, skipping (nothing to write bytecode to)"
-	exit 0
-fi
-
-# When running as a non-root user, site-packages is still root-owned and unwritable,
-# so this step would just fail loudly on every container start. Skip it.
-if [[ -n "${USER_IS_NON_ROOT}" ]]; then
-	echo "${log_prefix} USER_IS_NON_ROOT is set, skipping (site-packages is not writable)"
-	exit 0
-fi
-
-declare -r site_packages="$(python3 -c 'import site; print(site.getsitepackages()[0])')"
-
-# Deliberately scoped to packages that paperless.settings/paperless/__init__.py import
-# unconditionally on every manage.py invocation (Django itself, the always-loaded
-# INSTALLED_APPS, and celery). This is NOT "compile everything" - the optional AI stack
-# (torch, llama-index, sentence-transformers, ...) is intentionally excluded since it is
-# lazy-imported and large.
-declare -a scope=(
-	"${PAPERLESS_SRC_DIR}"
-	"${site_packages}/django"
-	"${site_packages}/celery"
-	"${site_packages}/kombu"
-	"${site_packages}/rest_framework"
-	"${site_packages}/django_filters"
-	"${site_packages}/whitenoise"
-	"${site_packages}/corsheaders"
-	"${site_packages}/django_extensions"
-	"${site_packages}/guardian"
-	"${site_packages}/allauth"
-	"${site_packages}/drf_spectacular"
-	"${site_packages}/drf_spectacular_sidecar"
-	"${site_packages}/treenode"
-	"${site_packages}/compression_middleware"
-)
-
-declare -a existing_scope=()
-for path in "${scope[@]}"; do
-	[[ -d "${path}" ]] && existing_scope+=("${path}")
-done
-
-echo "${log_prefix} Compiling bytecode for: ${existing_scope[*]}"
-declare -r start_seconds=${SECONDS}
-
-if ! PYTHONDONTWRITEBYTECODE= python3 -m compileall -q "${existing_scope[@]}"; then
-	echo "${log_prefix} WARNING: compileall reported errors (read-only filesystem or unwritable site-packages?); continuing without a bytecode cache"
-fi
-
-echo "${log_prefix} Done in $((SECONDS - start_seconds))s"

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/type

@@ -1 +0,0 @@
-oneshot

docker/rootfs/etc/s6-overlay/s6-rc.d/init-compile-bytecode/up

@@ -1 +0,0 @@
-/etc/s6-overlay/s6-rc.d/init-compile-bytecode/run

docs/configuration.md

@@ -1200,15 +1200,6 @@ still perform some basic text pre-processing before matching.
 
     Defaults to true, enabling the feature.
 
-#### [`PAPERLESS_CLASSIFIER_MATCH_THRESHOLD=<float>`](#PAPERLESS_CLASSIFIER_MATCH_THRESHOLD) {#PAPERLESS_CLASSIFIER_MATCH_THRESHOLD}
-
-: Sets the minimum confidence score (0.0-1.0) required for the automatic
-classifier to assign a correspondent, document type, or storage path to a
-document. Predictions below this threshold are discarded and the field is
-left unassigned, preventing low-confidence guesses from being applied.
-
-    Defaults to 0.6.
-
 #### [`PAPERLESS_DATE_PARSER_LANGUAGES=<lang>`](#PAPERLESS_DATE_PARSER_LANGUAGES) {#PAPERLESS_DATE_PARSER_LANGUAGES}
 
 : Specifies which language Paperless should use when parsing dates from documents.

src-ui/src/app/components/admin/settings/settings.component.html

@@ -112,22 +112,6 @@
 
                 <pngx-input-check i18n-title title="Use 'slim' sidebar (icons only)" formControlName="slimSidebarEnabled"></pngx-input-check>
 
-                <p class="mb-2 mt-3" i18n>Sidebar items to show:</p>
-                @for (option of sidebarItemOptions; track option.id) {
-                  <div class="form-check">
-                    <input
-                      class="form-check-input"
-                      type="checkbox"
-                      [id]="'sidebar-item-setting-' + option.id"
-                      [checked]="isSidebarItemShown(option.id)"
-                      (change)="toggleSidebarItem(option.id, $event.target.checked)"
-                    />
-                    <label class="form-check-label" [for]="'sidebar-item-setting-' + option.id">
-                      {{ option.label }}
-                    </label>
-                  </div>
-                }
-
               </div>
             </div>
 

src-ui/src/app/components/admin/settings/settings.component.spec.ts

@@ -24,7 +24,7 @@ import {
   SystemStatus,
   SystemStatusItemStatus,
 } from 'src/app/data/system-status'
-import { HideableSidebarItemID, SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { IfOwnerDirective } from 'src/app/directives/if-owner.directive'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { PermissionsGuard } from 'src/app/guards/permissions.guard'
@@ -209,45 +209,6 @@ describe('SettingsComponent', () => {
     fixture.detectChanges()
   }
 
-  it('supports configuring sidebar items and canceling changes', () => {
-    completeSetup()
-
-    component.toggleSidebarItem(HideableSidebarItemID.Workflows, false)
-    fixture.detectChanges()
-
-    expect(component.settingsForm.value.sidebarHiddenItems).toContain(
-      HideableSidebarItemID.Workflows
-    )
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Mail,
-      false
-    )
-
-    expect(component.settingsForm.value.sidebarHiddenItems).toContain(
-      HideableSidebarItemID.Mail
-    )
-
-    component.reset()
-
-    expect(component.settingsForm.value.sidebarHiddenItems).not.toContain(
-      HideableSidebarItemID.Workflows
-    )
-    expect(component.settingsForm.value.sidebarHiddenItems).not.toContain(
-      HideableSidebarItemID.Mail
-    )
-  })
-
-  it('enables sidebar item controls on general settings until destroyed', () => {
-    completeSetup()
-
-    expect(settingsService.organizingSidebarItems()).toBe(true)
-
-    component.ngOnDestroy()
-
-    expect(settingsService.organizingSidebarItems()).toBe(false)
-  })
-
   it('should support tabbed settings & change URL, prevent navigation if dirty confirmation rejected', async () => {
     completeSetup()
     const navigateSpy = jest.spyOn(router, 'navigate')
@@ -288,7 +249,6 @@ describe('SettingsComponent', () => {
 
   it('should support save local settings updating appearance settings and calling API, show error', () => {
     completeSetup()
-    component.toggleSidebarItem(HideableSidebarItemID.Workflows, false)
     const toastErrorSpy = jest.spyOn(toastService, 'showError')
     const toastSpy = jest.spyOn(toastService, 'show')
     const storeSpy = jest.spyOn(settingsService, 'storeSettings')
@@ -307,10 +267,7 @@ describe('SettingsComponent', () => {
     expect(toastErrorSpy).toHaveBeenCalled()
     expect(storeSpy).toHaveBeenCalled()
     expect(appearanceSettingsSpy).not.toHaveBeenCalled()
-    expect(setSpy).toHaveBeenCalledTimes(34)
+    expect(setSpy).toHaveBeenCalledTimes(33)
-    expect(setSpy).toHaveBeenCalledWith(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Workflows,
-    ])
 
     // succeed
     storeSpy.mockReturnValueOnce(of(true))

src-ui/src/app/components/admin/settings/settings.component.ts

@@ -39,12 +39,7 @@ import {
   SystemStatus,
   SystemStatusItemStatus,
 } from 'src/app/data/system-status'
-import {
+import { GlobalSearchType, SETTINGS_KEYS } from 'src/app/data/ui-settings'
-  GlobalSearchType,
-  HIDEABLE_SIDEBAR_ITEM_IDS,
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-} from 'src/app/data/ui-settings'
 import { User } from 'src/app/data/user'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { CustomDatePipe } from 'src/app/pipes/custom-date.pipe'
@@ -107,14 +102,6 @@ const documentDetailFieldOptions = [
   { id: DocumentDetailFieldID.Tags, label: $localize`Tags` },
 ]
 
-const sidebarItemLabels: Record<HideableSidebarItemID, string> = {
-  [HideableSidebarItemID.Dashboard]: $localize`Dashboard`,
-  [HideableSidebarItemID.SavedViews]: $localize`Saved Views`,
-  [HideableSidebarItemID.Workflows]: $localize`Workflows`,
-  [HideableSidebarItemID.Mail]: $localize`Mail`,
-  [HideableSidebarItemID.Documentation]: $localize`Documentation`,
-}
-
 @Component({
   selector: 'pngx-settings',
   templateUrl: './settings.component.html',
@@ -162,7 +149,6 @@ export class SettingsComponent
     bulkEditApplyOnClose: new FormControl(null),
     documentListItemPerPage: new FormControl(null),
     slimSidebarEnabled: new FormControl(null),
-    sidebarHiddenItems: new FormControl<HideableSidebarItemID[]>([]),
     darkModeUseSystem: new FormControl(null),
     darkModeEnabled: new FormControl(null),
     darkModeInvertThumbs: new FormControl(null),
@@ -200,7 +186,6 @@ export class SettingsComponent
 
   store: BehaviorSubject<any>
   storeSub: Subscription
-  sidebarItemsSub: Subscription
   isDirty$: Observable<boolean>
   isDirty: boolean = false
   unsubscribeNotifier: Subject<any> = new Subject()
@@ -218,10 +203,6 @@ export class SettingsComponent
   public readonly PdfEditorEditMode = PdfEditorEditMode
 
   public readonly documentDetailFieldOptions = documentDetailFieldOptions
-  public readonly sidebarItemOptions = HIDEABLE_SIDEBAR_ITEM_IDS.map((id) => ({
-    id,
-    label: sidebarItemLabels[id],
-  }))
 
   get systemStatusHasErrors(): boolean {
     const status = this.systemStatus()
@@ -249,10 +230,6 @@ export class SettingsComponent
 
   constructor() {
     super()
-    this.sidebarItemsSub =
-      this.settings.sidebarHiddenItemsEditingChanged.subscribe((hiddenItems) =>
-        this.settingsForm.controls.sidebarHiddenItems.setValue(hiddenItems)
-      )
     this.settings.settingsSaved.subscribe(() => {
       if (!this.savePending) this.initialize()
       this.savedViewsService.maybeRefreshDocumentCounts()
@@ -302,21 +279,14 @@ export class SettingsComponent
 
     this.activatedRoute.paramMap.subscribe((paramMap) => {
       const section = paramMap.get('section')
-      let navID = SettingsNavIDs.General
       if (section) {
         const navIDKey: string = Object.keys(SettingsNavIDs).find(
           (navID) => navID.toLowerCase() == section
         )
         if (navIDKey) {
-          navID = SettingsNavIDs[navIDKey]
+          this.activeNavID.set(SettingsNavIDs[navIDKey])
         }
       }
-      this.activeNavID.set(navID)
-      this.settings.sidebarHiddenItemsEditing.set(
-        navID === SettingsNavIDs.General
-          ? [...this.settingsForm.controls.sidebarHiddenItems.value]
-          : null
-      )
     })
   }
 
@@ -340,7 +310,6 @@ export class SettingsComponent
         SETTINGS_KEYS.DOCUMENT_LIST_SIZE
       ),
       slimSidebarEnabled: this.settings.get(SETTINGS_KEYS.SLIM_SIDEBAR),
-      sidebarHiddenItems: this.settings.get(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS),
       darkModeUseSystem: this.settings.get(SETTINGS_KEYS.DARK_MODE_USE_SYSTEM),
       darkModeEnabled: this.settings.get(SETTINGS_KEYS.DARK_MODE_ENABLED),
       darkModeInvertThumbs: this.settings.get(
@@ -467,12 +436,6 @@ export class SettingsComponent
       this.settingsForm.patchValue(currentFormValue)
     }
 
-    if (this.settings.organizingSidebarItems()) {
-      this.settings.sidebarHiddenItemsEditing.set([
-        ...this.settingsForm.controls.sidebarHiddenItems.value,
-      ])
-    }
-
     if (this.canViewSystemStatus) {
       this.systemStatusService.get().subscribe((status) => {
         this.systemStatus.set(status)
@@ -481,18 +444,8 @@ export class SettingsComponent
   }
 
   ngOnDestroy() {
-    this.settings.sidebarHiddenItemsEditing.set(null)
     if (this.isDirty) this.settings.updateAppearanceSettings() // in case user changed appearance but didn't save
     this.storeSub && this.storeSub.unsubscribe()
-    this.sidebarItemsSub.unsubscribe()
-  }
-
-  isSidebarItemShown(item: HideableSidebarItemID): boolean {
-    return !(this.settingsForm.value.sidebarHiddenItems || []).includes(item)
-  }
-
-  toggleSidebarItem(item: HideableSidebarItemID, checked: boolean): void {
-    this.settings.updateSidebarItemVisibility(item, checked)
   }
 
   public saveSettings() {
@@ -520,10 +473,6 @@ export class SettingsComponent
       SETTINGS_KEYS.SLIM_SIDEBAR,
       this.settingsForm.value.slimSidebarEnabled
     )
-    this.settings.set(
-      SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS,
-      this.settingsForm.value.sidebarHiddenItems
-    )
     this.settings.set(
       SETTINGS_KEYS.DARK_MODE_USE_SYSTEM,
       this.settingsForm.value.darkModeUseSystem
@@ -683,11 +632,6 @@ export class SettingsComponent
 
   reset() {
     this.settingsForm.patchValue(this.store.getValue())
-    if (this.settings.organizingSidebarItems()) {
-      this.settings.sidebarHiddenItemsEditing.set([
-        ...this.settingsForm.controls.sidebarHiddenItems.value,
-      ])
-    }
   }
 
   clearThemeColor() {

src-ui/src/app/components/app-frame/app-frame.component.html

@@ -86,15 +86,12 @@
       }
       <div class="sidebar-sticky pt-3 pb-1 d-flex flex-column justify-space-around">
         <ul class="nav flex-column">
-          <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard) && !settingsService.organizingSidebarItems()">
+          <li class="nav-item app-link">
-            <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="dashboard" routerLinkActive="active" (click)="closeMenu()"
+            <a class="nav-link" routerLink="dashboard" routerLinkActive="active" (click)="closeMenu()"
               ngbPopover="Dashboard" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
               container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
               <i-bs class="me-2" name="house"></i-bs><span class="nav-link-label"><ng-container i18n>Dashboard</ng-container></span>
             </a>
-            @if (settingsService.organizingSidebarItems()) {
-              <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Dashboard" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Dashboard)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Dashboard, $event)"></pngx-input-switch>
-            }
           </li>
           <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.Document }">
             <a class="nav-link" routerLink="documents" routerLinkActive="active"
@@ -240,38 +237,29 @@
                 </div>
               </li>
             }
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews) && !settingsService.organizingSidebarItems()" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.SavedView }">
+            <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.SavedView }">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="savedviews" routerLinkActive="active" (click)="closeMenu()"
+              <a class="nav-link" routerLink="savedviews" routerLinkActive="active" (click)="closeMenu()"
                 ngbPopover="Saved Views" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
                 container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="window-stack"></i-bs><span class="nav-link-label"><ng-container i18n>Saved Views</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Saved Views" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.SavedViews)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.SavedViews, $event)"></pngx-input-switch>
-              }
             </li>
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows) && !settingsService.organizingSidebarItems()"
+            <li class="nav-item app-link"
               *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.Workflow }"
               tourAnchor="tour.workflows">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="workflows" routerLinkActive="active" (click)="closeMenu()"
+              <a class="nav-link" routerLink="workflows" routerLinkActive="active" (click)="closeMenu()"
                 ngbPopover="Workflows" i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end"
                 container="body" triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="boxes"></i-bs><span class="nav-link-label"><ng-container i18n>Workflows</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Workflows" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Workflows, $event)"></pngx-input-switch>
-              }
             </li>
-            <li class="nav-item app-link position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail) && !settingsService.organizingSidebarItems()" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.MailAccount }"
+            <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.View, type: PermissionType.MailAccount }"
               tourAnchor="tour.mail">
-              <a class="nav-link" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()" routerLink="mail" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Mail"
+              <a class="nav-link" routerLink="mail" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Mail"
                 i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end" container="body"
                 triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="me-2" name="envelope"></i-bs><span class="nav-link-label"><ng-container i18n>Mail</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Mail" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Mail)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Mail, $event)"></pngx-input-switch>
-              }
             </li>
             <li class="nav-item app-link" *pngxIfPermissions="{ action: PermissionAction.Delete, type: PermissionType.Document }">
               <a class="nav-link" routerLink="trash" routerLinkActive="active" (click)="closeMenu()" ngbPopover="Trash"
@@ -334,16 +322,13 @@
                 </a>
               </li>
             }
-            <li class="nav-item mt-2 position-relative" [class.d-none]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation) && !settingsService.organizingSidebarItems()" tourAnchor="tour.outro">
+            <li class="nav-item mt-2" tourAnchor="tour.outro">
-              <a class="text-muted small d-flex align-items-center flex-wrap text-decoration-none nav-anchor" [class.opacity-50]="settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation)" [class.pe-5]="settingsService.organizingSidebarItems() && !slimSidebarEnabled && !slimSidebarAnimating()"
+              <a class="text-muted small d-flex align-items-center flex-wrap text-decoration-none nav-anchor"
                 target="_blank" rel="noopener noreferrer" href="https://docs.paperless-ngx.com" ngbPopover="Documentation"
                 i18n-ngbPopover [disablePopover]="!slimSidebarPopoversEnabled" placement="end" container="body"
                 triggers="mouseenter:mouseleave" popoverClass="popover-slim">
                 <i-bs class="d-flex me-2" name="question-circle"></i-bs><span><ng-container i18n>Documentation</ng-container></span>
               </a>
-              @if (settingsService.organizingSidebarItems()) {
-                <pngx-input-switch class="position-absolute top-50 end-0 translate-middle-y me-1" [class.d-none]="slimSidebarEnabled || slimSidebarAnimating()" [compact]="true" title="Documentation" i18n-title [ngModel]="!settingsService.sidebarItemIsHidden(HideableSidebarItemID.Documentation)" (ngModelChange)="toggleSidebarItem(HideableSidebarItemID.Documentation, $event)"></pngx-input-switch>
-              }
             </li>
             <li class="nav-item" [class.visually-hidden]="slimSidebarEnabled">
               <div class="text-muted small d-flex align-items-center flex-wrap nav-label">

src-ui/src/app/components/app-frame/app-frame.component.spec.ts

@@ -15,7 +15,7 @@ import { provideUiTour } from 'ngx-ui-tour-ng-bootstrap'
 import { of, throwError } from 'rxjs'
 import { routes } from 'src/app/app-routing.module'
 import { SavedView } from 'src/app/data/saved-view'
-import { HideableSidebarItemID, SETTINGS_KEYS } from 'src/app/data/ui-settings'
+import { SETTINGS_KEYS } from 'src/app/data/ui-settings'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { PermissionsGuard } from 'src/app/guards/permissions.guard'
 import {
@@ -287,82 +287,6 @@ describe('AppFrameComponent', () => {
     jest.useRealTimers()
   })
 
-  it('should hide configured sidebar items', () => {
-    settingsService.set(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Dashboard,
-      HideableSidebarItemID.Workflows,
-    ])
-    fixture.detectChanges()
-
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]')
-        .parentElement.classList
-    ).toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="workflows"]')
-        .parentElement.classList
-    ).toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="mail"]').parentElement
-        .classList
-    ).not.toContain('d-none')
-  })
-
-  it('should show hidden items and visibility switches while customizing', () => {
-    settingsService.set(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS, [
-      HideableSidebarItemID.Dashboard,
-    ])
-    settingsService.sidebarHiddenItemsEditing.set([
-      HideableSidebarItemID.Dashboard,
-    ])
-    fixture.detectChanges()
-
-    expect(
-      fixture.nativeElement.querySelectorAll('pngx-input-switch').length
-    ).toBe(5)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]')
-        .parentElement.classList
-    ).not.toContain('d-none')
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).toContain('opacity-50')
-
-    settingsService.set(SETTINGS_KEYS.SLIM_SIDEBAR, true)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => toggle.classList.contains('d-none'))
-    ).toBe(true)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).not.toContain('pe-5')
-
-    settingsService.set(SETTINGS_KEYS.SLIM_SIDEBAR, false)
-    component.slimSidebarAnimating.set(true)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => toggle.classList.contains('d-none'))
-    ).toBe(true)
-
-    component.slimSidebarAnimating.set(false)
-    fixture.detectChanges()
-
-    expect(
-      Array.from(
-        fixture.nativeElement.querySelectorAll('pngx-input-switch')
-      ).every((toggle: HTMLElement) => !toggle.classList.contains('d-none'))
-    ).toBe(true)
-    expect(
-      fixture.nativeElement.querySelector('[routerLink="dashboard"]').classList
-    ).toContain('pe-5')
-  })
-
   it('should show error on toggle slim sidebar if store settings fails', () => {
     jest.spyOn(console, 'warn').mockImplementation(() => {})
     const toastSpy = jest.spyOn(toastService, 'showError')

src-ui/src/app/components/app-frame/app-frame.component.ts

@@ -7,7 +7,6 @@ import {
 } from '@angular/cdk/drag-drop'
 import { NgClass } from '@angular/common'
 import { Component, HostListener, inject, OnInit, signal } from '@angular/core'
-import { FormsModule } from '@angular/forms'
 import { ActivatedRoute, Router, RouterModule } from '@angular/router'
 import {
   NgbCollapseModule,
@@ -22,11 +21,7 @@ import { Observable } from 'rxjs'
 import { first } from 'rxjs/operators'
 import { Document } from 'src/app/data/document'
 import { SavedView } from 'src/app/data/saved-view'
-import {
+import { CollapsibleSection, SETTINGS_KEYS } from 'src/app/data/ui-settings'
-  CollapsibleSection,
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-} from 'src/app/data/ui-settings'
 import { IfPermissionsDirective } from 'src/app/directives/if-permissions.directive'
 import { ComponentCanDeactivate } from 'src/app/guards/dirty-doc.guard'
 import { DocumentTitlePipe } from 'src/app/pipes/document-title.pipe'
@@ -53,7 +48,6 @@ import { ChatComponent } from '../chat/chat/chat.component'
 import { BrandMarkComponent } from '../common/logo/brand-mark/brand-mark.component'
 import { LogoComponent } from '../common/logo/logo.component'
 import { ProfileEditDialogComponent } from '../common/profile-edit-dialog/profile-edit-dialog.component'
-import { SwitchComponent } from '../common/input/switch/switch.component'
 import { DocumentDetailComponent } from '../document-detail/document-detail.component'
 import { ComponentWithPermissions } from '../with-permissions/with-permissions.component'
 import { GlobalSearchComponent } from './global-search/global-search.component'
@@ -82,8 +76,6 @@ const SCROLL_THRESHOLD = 16
     NgxBootstrapIconsModule,
     DragDropModule,
     TourNgBootstrap,
-    FormsModule,
-    SwitchComponent,
   ],
 })
 export class AppFrameComponent
@@ -106,7 +98,6 @@ export class AppFrameComponent
   readonly isMenuCollapsed = signal(true)
   readonly slimSidebarAnimating = signal(false)
   readonly mobileSearchHidden = signal(false)
-  readonly HideableSidebarItemID = HideableSidebarItemID
   private readonly versionSetting = this.settingsService.getSignal<string>(
     SETTINGS_KEYS.VERSION
   )
@@ -204,10 +195,6 @@ export class AppFrameComponent
     }, 200) // slightly longer than css animation for slim sidebar
   }
 
-  toggleSidebarItem(item: HideableSidebarItemID, visible: boolean): void {
-    this.settingsService.updateSidebarItemVisibility(item, visible)
-  }
-
   toggleAttributesSections(event?: Event): void {
     event?.preventDefault()
     event?.stopPropagation()

src-ui/src/app/components/common/input/switch/switch.component.html

@@ -1,6 +1,6 @@
-<div [class.mb-3]="!compact">
+<div class="mb-3">
-  <div [class.row]="!compact">
+  <div class="row">
-    @if (!horizontal && !compact) {
+    @if (!horizontal) {
       <div class="d-flex align-items-center position-relative hidden-button-container col-md-3">
         <label class="form-label" [for]="inputId" [ngbTooltip]="showUnsetNote && isUnset ? tipContent: null" placement="end">
           {{title}}
@@ -17,8 +17,8 @@
     }
     <div [ngClass]="{'align-items-center': horizontal, 'd-flex': horizontal}">
       <div class="form-check form-switch">
-        <input #inputField type="checkbox" class="form-check-input" [id]="inputId" [(ngModel)]="value" [ngModelOptions]="{standalone: true}" (change)="onChange(value)" (blur)="onTouched()" [disabled]="disabled" [attr.aria-label]="compact ? title : null">
+        <input #inputField type="checkbox" class="form-check-input" [id]="inputId" [(ngModel)]="value" [ngModelOptions]="{standalone: true}" (change)="onChange(value)" (blur)="onTouched()" [disabled]="disabled">
-        @if (horizontal && !compact) {
+        @if (horizontal) {
           <label class="form-check-label" [class.text-muted]="showUnsetNote && isUnset" [for]="inputId" [ngbTooltip]="showUnsetNote && isUnset ? tipContent: null" placement="end">
             {{title}}
             @if (showUnsetNote && isUnset) {

src-ui/src/app/components/common/input/switch/switch.component.spec.ts

@@ -48,14 +48,4 @@ describe('SwitchComponent', () => {
     component.value = undefined
     expect(component.isUnset).toBeTruthy()
   })
-
-  it('should support a compact layout', () => {
-    component.compact = true
-    component.title = 'Test switch'
-    fixture.detectChanges()
-
-    expect(fixture.nativeElement.querySelector('.mb-3')).toBeNull()
-    expect(fixture.nativeElement.querySelector('.row')).toBeNull()
-    expect(input.getAttribute('aria-label')).toEqual('Test switch')
-  })
 })

src-ui/src/app/components/common/input/switch/switch.component.ts

@@ -25,9 +25,6 @@ export class SwitchComponent extends AbstractInputComponent<boolean> {
   @Input()
   showUnsetNote: boolean = false
 
-  @Input()
-  compact: boolean = false
-
   constructor() {
     super()
   }

src-ui/src/app/data/ui-settings.ts

@@ -24,16 +24,6 @@ export enum CollapsibleSection {
   ATTRIBUTES = 'attributes',
 }
 
-export enum HideableSidebarItemID {
-  Dashboard = 'dashboard',
-  SavedViews = 'saved_views',
-  Workflows = 'workflows',
-  Mail = 'mail',
-  Documentation = 'documentation',
-}
-
-export const HIDEABLE_SIDEBAR_ITEM_IDS = Object.values(HideableSidebarItemID)
-
 export const PAPERLESS_GREEN_HEX = '#17541f'
 
 export const SETTINGS_KEYS = {
@@ -66,7 +56,6 @@ export const SETTINGS_KEYS = {
   NOTES_ENABLED: 'general-settings:notes-enabled',
   AUDITLOG_ENABLED: 'general-settings:auditlog-enabled',
   SLIM_SIDEBAR: 'general-settings:slim-sidebar',
-  SIDEBAR_HIDDEN_ITEMS: 'general-settings:sidebar:hidden-items',
   ATTRIBUTES_SECTIONS_COLLAPSED:
     'general-settings:attributes-sections-collapsed',
   UPDATE_CHECKING_ENABLED: 'general-settings:update-checking:enabled',
@@ -138,11 +127,6 @@ export const SETTINGS: UiSetting[] = [
     type: 'boolean',
     default: false,
   },
-  {
-    key: SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS,
-    type: 'array',
-    default: [],
-  },
   {
     key: SETTINGS_KEYS.ATTRIBUTES_SECTIONS_COLLAPSED,
     type: 'array',

src-ui/src/app/services/settings.service.spec.ts

@@ -14,11 +14,7 @@ import { CustomFieldDataType } from '../data/custom-field'
 import { DEFAULT_DISPLAY_FIELDS, DisplayField } from '../data/document'
 import { SavedView } from '../data/saved-view'
 import { RemoteOCRModeConfig } from '../data/paperless-config'
-import {
+import { SETTINGS_KEYS, UiSettings } from '../data/ui-settings'
-  HideableSidebarItemID,
-  SETTINGS_KEYS,
-  UiSettings,
-} from '../data/ui-settings'
 import { PermissionsService } from './permissions.service'
 import { CustomFieldsService } from './rest/custom-fields.service'
 import { SettingsService } from './settings.service'
@@ -234,35 +230,6 @@ describe('SettingsService', () => {
     expect(notesEnabled()).toBeFalsy()
   })
 
-  it('updates sidebar item visibility', () => {
-    httpTestingController
-      .expectOne(`${environment.apiBaseUrl}ui_settings/`)
-      .flush(ui_settings)
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(false)
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Workflows,
-      false
-    )
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(true)
-    expect(settingsService.get(SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS)).toEqual([])
-
-    settingsService.updateSidebarItemVisibility(
-      HideableSidebarItemID.Workflows,
-      true
-    )
-
-    expect(
-      settingsService.sidebarItemIsHidden(HideableSidebarItemID.Workflows)
-    ).toBe(false)
-  })
-
   it('updates setting signals when settings are reinitialized', () => {
     let req = httpTestingController.expectOne(
       `${environment.apiBaseUrl}ui_settings/`

src-ui/src/app/services/settings.service.ts

@@ -24,7 +24,6 @@ import { DEFAULT_DISPLAY_FIELDS, DisplayField } from '../data/document'
 import { RemoteOCRModeConfig } from '../data/paperless-config'
 import { SavedView } from '../data/saved-view'
 import {
-  HideableSidebarItemID,
   PAPERLESS_GREEN_HEX,
   SETTINGS,
   SETTINGS_KEYS,
@@ -314,18 +313,6 @@ export class SettingsService {
   readonly globalDropzoneEnabled = signal(true)
   readonly globalDropzoneActive = signal(false)
   readonly organizingSidebarSavedViews = signal(false)
-  readonly sidebarHiddenItemsEditing = signal<HideableSidebarItemID[] | null>(
-    null
-  )
-  readonly organizingSidebarItems = computed(
-    () => this.sidebarHiddenItemsEditing() !== null
-  )
-  readonly sidebarHiddenItemsEditingChanged = new EventEmitter<
-    HideableSidebarItemID[]
-  >()
-  readonly hiddenSidebarItems = this.getSignal<HideableSidebarItemID[]>(
-    SETTINGS_KEYS.SIDEBAR_HIDDEN_ITEMS
-  )
 
   readonly allDisplayFields = signal<Array<{ id: DisplayField; name: string }>>(
     DEFAULT_DISPLAY_FIELDS
@@ -762,29 +749,6 @@ export class SettingsService {
     return this.storeSettings()
   }
 
-  sidebarItemIsHidden(item: HideableSidebarItemID): boolean {
-    return (
-      this.sidebarHiddenItemsEditing() ?? this.hiddenSidebarItems()
-    ).includes(item)
-  }
-
-  updateSidebarItemVisibility(
-    item: HideableSidebarItemID,
-    visible: boolean
-  ): void {
-    const hiddenItems = new Set(
-      this.sidebarHiddenItemsEditing() ?? this.hiddenSidebarItems()
-    )
-    if (visible) {
-      hiddenItems.delete(item)
-    } else {
-      hiddenItems.add(item)
-    }
-    const updatedHiddenItems = [...hiddenItems]
-    this.sidebarHiddenItemsEditing.set(updatedHiddenItems)
-    this.sidebarHiddenItemsEditingChanged.emit(updatedHiddenItems)
-  }
-
   updateSavedViewsVisibility(
     dashboardVisibleViewIds: number[],
     sidebarVisibleViewIds: number[]

src/documents/apps.py

@@ -28,7 +28,6 @@ class DocumentsConfig(AppConfig):
         document_consumption_finished.connect(set_storage_path)
         document_consumption_finished.connect(add_to_index)
         document_consumption_finished.connect(run_workflows_added)
-        document_consumption_finished.connect(add_to_index)
         document_consumption_finished.connect(add_or_update_document_in_llm_index)
         document_updated.connect(run_workflows_updated)
         document_updated.connect(send_websocket_document_updated)

src/documents/bulk_edit.py

@@ -2,7 +2,6 @@ from __future__ import annotations
 
 import logging
 import tempfile
-import uuid
 from pathlib import Path
 from typing import TYPE_CHECKING
 from typing import Literal
@@ -28,7 +27,7 @@ from documents.models import DocumentType
 from documents.models import PaperlessTask
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import set_permissions_for_objects
+from documents.permissions import set_permissions_for_object
 from documents.plugins.helpers import DocumentsStatusManager
 from documents.tasks import bulk_update_documents
 from documents.tasks import consume_file
@@ -299,55 +298,53 @@ def modify_custom_fields(
 ) -> Literal["OK"]:
     qs = Document.objects.filter(id__in=doc_ids).only("pk")
     affected_docs = list(qs.values_list("pk", flat=True))
-    # Ensure add_custom_fields is a list of (int, value) tuples, supports old API
+    # Ensure add_custom_fields is a list of tuples, supports old API
     add_custom_fields = (
-        [(int(field), value) for field, value in add_custom_fields.items()]
+        add_custom_fields.items()
         if isinstance(add_custom_fields, dict)
-        else [(int(field), None) for field in add_custom_fields]
+        else [(field, None) for field in add_custom_fields]
     )
 
-    # Resolved once, instead of re-querying the same field for every document
+    custom_fields = CustomField.objects.filter(
-    custom_fields_by_id: dict[int, CustomField] = CustomField.objects.in_bulk(
+        id__in=[int(field) for field, _ in add_custom_fields],
-        [field_id for field_id, _ in add_custom_fields],
+    ).distinct()
-    )
-    # Passed to update_or_create() below rather than a bare id, so the FK is
-    # cached on the created instance and auditlog's post_save receiver does
-    # not reload it per row. Only needed for additions. content is deferred:
-    # the one field here that is both large and unused.
-    docs_by_id: dict[int, Document] = (
-        Document.objects.defer("content").in_bulk(affected_docs)
-        if add_custom_fields
-        else {}
-    )
     for field_id, value in add_custom_fields:
-        custom_field = custom_fields_by_id[field_id]
-        value_field = CustomFieldInstance.TYPE_TO_DATA_STORE_NAME_MAP[
-            custom_field.data_type
-        ]
-        is_doclink = custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK
         for doc_id in affected_docs:
-            if is_doclink and value and doc_id in value:
+            defaults = {}
-                # Prevent self-linking
+            custom_field = custom_fields.get(id=field_id)
-                continue
+            if custom_field:
+                value_field = CustomFieldInstance.TYPE_TO_DATA_STORE_NAME_MAP[
+                    custom_field.data_type
+                ]
+                defaults[value_field] = value
+                if (
+                    custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK
+                    and value
+                    and doc_id in value
+                ):
+                    # Prevent self-linking
+                    continue
             CustomFieldInstance.objects.update_or_create(
-                document=docs_by_id[doc_id],
+                document_id=doc_id,
-                field=custom_field,
+                field_id=field_id,
-                defaults={value_field: value},
+                defaults=defaults,
             )
-            if is_doclink:
+            if custom_field.data_type == CustomField.FieldDataType.DOCUMENTLINK:
-                reflect_doclinks(docs_by_id[doc_id], custom_field, value)
+                doc = Document.objects.get(id=doc_id)
+                reflect_doclinks(doc, custom_field, value)
 
-    # For doc link fields that are being removed, remove symmetrical links.
+    # For doc link fields that are being removed, remove symmetrical links
-    # select_related avoids a per-instance reload of the document and field.
     for doclink_being_removed_instance in CustomFieldInstance.objects.filter(
         document_id__in=affected_docs,
         field__id__in=remove_custom_fields,
         field__data_type=CustomField.FieldDataType.DOCUMENTLINK,
         value_document_ids__isnull=False,
-    ).select_related("field", "document"):
+    ):
         for target_doc_id in doclink_being_removed_instance.value:
             remove_doclink(
-                document=doclink_being_removed_instance.document,
+                document=Document.objects.get(
+                    id=doclink_being_removed_instance.document.id,
+                ),
                 field=doclink_being_removed_instance.field,
                 target_doc_id=target_doc_id,
             )
@@ -382,7 +379,7 @@ def delete(doc_ids: list[int]) -> Literal["OK"]:
         )
         delete_ids = list({*doc_ids, *version_ids})
 
-        Document.objects.filter(id__in=delete_ids).delete(transaction_id=uuid.uuid4())
+        Document.objects.filter(id__in=delete_ids).delete()
 
         from documents.search import get_backend
 
@@ -433,13 +430,10 @@ def set_permissions(
     else:
         qs.update(owner=owner)
 
+    for doc in qs:
+        set_permissions_for_object(permissions=set_permissions, object=doc, merge=merge)
+
     affected_docs = list(qs.values_list("pk", flat=True))
-    set_permissions_for_objects(
-        permissions=set_permissions,
-        model=Document,
-        pks=affected_docs,
-        merge=merge,
-    )
 
     bulk_update_documents.apply_async(
         kwargs={"document_ids": affected_docs},
@@ -1183,13 +1177,10 @@ def remove_doclink(
     """
     Removes a 'symmetrical' link to `document` from the target document's existing custom field instance
     """
-    # select_related: a signal receiver (auditlog) touches .document/.field on
+    target_doc_field_instance = CustomFieldInstance.objects.filter(
-    # the save() below, without this that is a per-call reload query
+        document_id=target_doc_id,
-    target_doc_field_instance = (
+        field=field,
-        CustomFieldInstance.objects.filter(document_id=target_doc_id, field=field)
+    ).first()
-        .select_related("document", "field")
-        .first()
-    )
     if (
         target_doc_field_instance is not None
         and document.id in target_doc_field_instance.value

src/documents/classifier.py

@@ -34,27 +34,6 @@ from paperless.signed_pickle import signed_pickle_loads
 
 logger = logging.getLogger("paperless.classifier")
 
-
-def _predict_with_threshold(classifier, X, threshold: float) -> int | None:
-    """
-    Return the predicted class id, or None if:
-    - the prediction is -1 (no match), or
-    - the winning class probability is below the configured threshold.
-
-    Using predict_proba() instead of predict() lets us apply a minimum-confidence
-    cutoff so that uncertain predictions are discarded rather than assigned.
-    """
-    probas = classifier.predict_proba(X)[0]
-    best_idx = int(probas.argmax())
-    best_class = int(classifier.classes_[best_idx])
-
-    if best_class == -1:
-        return None
-    if threshold > 0.0 and probas[best_idx] < threshold:
-        return None
-    return best_class
-
-
 ADVANCED_TEXT_PROCESSING_ENABLED = (
     settings.NLTK_LANGUAGE is not None and settings.NLTK_ENABLED
 )
@@ -123,8 +102,7 @@ class DocumentClassifier:
     # v8 - Added storage path classifier
     # v9 - Changed from hashing to time/ids for re-train check
     # v10 - HMAC-signed model file
-    # v11 - Use sample_weight for balanced training; predict_proba with threshold
+    FORMAT_VERSION = 10
-    FORMAT_VERSION = 11
 
     HMAC_SIZE = 32  # SHA-256 digest length
 
@@ -346,13 +324,6 @@ class DocumentClassifier:
         from sklearn.preprocessing import LabelBinarizer
         from sklearn.preprocessing import MultiLabelBinarizer
 
-        # MLPClassifier does not support class_weight directly
-        # (https://github.com/scikit-learn/scikit-learn/issues/9113), so we use
-        # compute_sample_weight to balance classes during training and prevent
-        # over-represented correspondents from dominating predictions.
-        # https://scikit-learn.org/stable/modules/generated/sklearn.utils.class_weight.compute_sample_weight.html
-        from sklearn.utils.class_weight import compute_sample_weight
-
         # Step 2: vectorize data
         logger.debug("Vectorizing data...")
         notify("Vectorizing document content...")
@@ -398,7 +369,7 @@ class DocumentClassifier:
                 self.tags_binarizer = MultiLabelBinarizer()
                 labels_tags_vectorized = self.tags_binarizer.fit_transform(labels_tags)
 
-            self.tags_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.tags_classifier = MLPClassifier(tol=0.01)
             self.tags_classifier.fit(data_vectorized, labels_tags_vectorized)
         else:
             self.tags_classifier = None
@@ -409,12 +380,8 @@ class DocumentClassifier:
             notify(
                 f"Training correspondent classifier ({num_correspondents} correspondent(s))...",
             )
-            self.correspondent_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.correspondent_classifier = MLPClassifier(tol=0.01)
-            self.correspondent_classifier.fit(
+            self.correspondent_classifier.fit(data_vectorized, labels_correspondent)
-                data_vectorized,
-                labels_correspondent,
-                sample_weight=compute_sample_weight("balanced", labels_correspondent),
-            )
         else:
             self.correspondent_classifier = None
             logger.debug(
@@ -426,12 +393,8 @@ class DocumentClassifier:
             notify(
                 f"Training document type classifier ({num_document_types} type(s))...",
             )
-            self.document_type_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.document_type_classifier = MLPClassifier(tol=0.01)
-            self.document_type_classifier.fit(
+            self.document_type_classifier.fit(data_vectorized, labels_document_type)
-                data_vectorized,
-                labels_document_type,
-                sample_weight=compute_sample_weight("balanced", labels_document_type),
-            )
         else:
             self.document_type_classifier = None
             logger.debug(
@@ -443,11 +406,10 @@ class DocumentClassifier:
                 "Training storage paths classifier...",
             )
             notify(f"Training storage path classifier ({num_storage_paths} path(s))...")
-            self.storage_path_classifier = MLPClassifier(tol=0.01, random_state=0)
+            self.storage_path_classifier = MLPClassifier(tol=0.01)
             self.storage_path_classifier.fit(
                 data_vectorized,
                 labels_storage_path,
-                sample_weight=compute_sample_weight("balanced", labels_storage_path),
             )
         else:
             self.storage_path_classifier = None
@@ -584,24 +546,24 @@ class DocumentClassifier:
     def predict_correspondent(self, content: str) -> int | None:
         if self.correspondent_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
+            correspondent_id = self.correspondent_classifier.predict(X)
-                self.correspondent_classifier,
+            if correspondent_id != -1:
-                X,
+                return correspondent_id
-                settings.CLASSIFIER_MATCH_THRESHOLD,
+            else:
-            )
+                return None
-            return predicted_id
+        else:
-        return None
+            return None
 
     def predict_document_type(self, content: str) -> int | None:
         if self.document_type_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
+            document_type_id = self.document_type_classifier.predict(X)
-                self.document_type_classifier,
+            if document_type_id != -1:
-                X,
+                return document_type_id
-                settings.CLASSIFIER_MATCH_THRESHOLD,
+            else:
-            )
+                return None
-            return predicted_id
+        else:
-        return None
+            return None
 
     def predict_tags(self, content: str) -> list[int]:
         from sklearn.utils.multiclass import type_of_target
@@ -627,10 +589,10 @@ class DocumentClassifier:
     def predict_storage_path(self, content: str) -> int | None:
         if self.storage_path_classifier:
             X = self._vectorize(content)
-            predicted_id = _predict_with_threshold(
+            storage_path_id = self.storage_path_classifier.predict(X)
-                self.storage_path_classifier,
+            if storage_path_id != -1:
-                X,
+                return storage_path_id
-                settings.CLASSIFIER_MATCH_THRESHOLD,
+            else:
-            )
+                return None
-            return predicted_id
+        else:
-        return None
+            return None

src/documents/consumer.py

@@ -52,6 +52,7 @@ from documents.templating.workflows import parse_w_workflow_placeholders
 from documents.utils import compute_checksum
 from documents.utils import copy_basic_file_stats
 from documents.utils import copy_file_with_basic_stats
+from documents.utils import normalize_unicode
 from documents.utils import run_subprocess
 from paperless.config import OcrConfig
 from paperless.config import RemoteOCRConfig
@@ -201,7 +202,9 @@ class ConsumerPluginMixin:
 
         self.renew_logging_group()
 
-        self.filename = self.metadata.filename or self.input_doc.original_file.name
+        self.filename = normalize_unicode(
+            self.metadata.filename or self.input_doc.original_file.name,
+        )
 
     def _send_progress(
         self,

src/documents/data_models.py

@@ -129,7 +129,7 @@ class DocumentMetadataOverrides:
         )
         overrides.custom_fields = {
             custom_field.field.id: custom_field.value
-            for custom_field in doc.custom_fields.select_related("field").all()
+            for custom_field in doc.custom_fields.all()
         }
 
         groups_with_perms = get_groups_with_perms(

src/documents/management/commands/document_consumer.py

@@ -156,15 +156,6 @@ class FileStabilityTracker:
                     logger.debug(f"File disappeared during stability check: {path}")
                 continue
 
-            # Stable, but empty: some scanners create a zero byte placeholder
-            # and only write the page some time later. Consuming it now can
-            # only fail  so drop it and let the writer's next event
-            # (or the periodic rescan) bring it back once it has content
-            if not tracked.last_size:
-                to_remove.append(path)
-                logger.debug("Ignoring stable but empty file: %s", path)
-                continue
-
             # File is stable, we can return it
             to_yield.append(path)
             logger.info(f"File is stable: {path}")

src/documents/matching.py

@@ -21,6 +21,7 @@ from documents.models import Workflow
 from documents.models import WorkflowTrigger
 from documents.permissions import permitted_object_ids
 from documents.regex import safe_regex_search
+from documents.utils import normalize_unicode
 
 if TYPE_CHECKING:
     from django.db.models import QuerySet
@@ -311,11 +312,12 @@ def consumable_document_matches_workflow(
         trigger_matched = False
 
     # Document filename vs trigger filename
+    document_filename = normalize_unicode(document.original_file.name)
     if (
         trigger.filter_filename is not None
         and len(trigger.filter_filename) > 0
         and not fnmatch(
-            document.original_file.name.lower(),
+            document_filename.lower(),
             trigger.filter_filename.lower(),
         )
     ):
@@ -328,10 +330,12 @@ def consumable_document_matches_workflow(
     # Document path vs trigger path
 
     # Use the original_path if set, else us the original_file
-    match_against = (
+    match_against = normalize_unicode(
-        document.original_path
+        str(
-        if document.original_path is not None
+            document.original_path
-        else document.original_file
+            if document.original_path is not None
+            else document.original_file,
+        ),
     )
 
     if (
@@ -536,7 +540,7 @@ def existing_document_matches_workflow(
         and len(trigger.filter_filename) > 0
         and document.original_filename is not None
         and not fnmatch(
-            document.original_filename.lower(),
+            normalize_unicode(document.original_filename).lower(),
             trigger.filter_filename.lower(),
         )
     ):

src/documents/models.py

@@ -1,5 +1,4 @@
 import datetime
-import uuid
 from pathlib import Path
 from typing import Final
 
@@ -28,6 +27,7 @@ from django_softdelete.models import SoftDeleteModel
 
 from documents.data_models import DocumentSource
 from documents.parsers import get_default_file_extension
+from documents.utils import normalize_unicode
 
 
 class ModelWithOwner(models.Model):
@@ -375,7 +375,6 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         If the queryset already annotated ``effective_content``, that value is used.
         """
         # Here to avoid circular import
-        from documents.versioning import LATEST_VERSION_CONTENT_PREFETCH_ATTR
         from documents.versioning import sort_versions_newest_first
         from documents.versioning import versions_newest_first
 
@@ -385,19 +384,6 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         if self.root_document_id is not None or self.pk is None:
             return self.content
 
-        latest_version_prefetch = getattr(
-            self,
-            LATEST_VERSION_CONTENT_PREFETCH_ATTR,
-            None,
-        )
-        if latest_version_prefetch is not None:
-            # Empty list means prefetch ran and found no versions — use own content.
-            return (
-                latest_version_prefetch[0].content
-                if latest_version_prefetch
-                else self.content
-            )
-
         prefetched_cache = getattr(self, "_prefetched_objects_cache", None)
         prefetched_versions = (
             prefetched_cache.get("versions")
@@ -482,7 +468,7 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
         context_document = (
             self.root_document if self.root_document_id is not None else self
         )
-        result = str(context_document)
+        result = normalize_unicode(str(context_document))
 
         if counter:
             result += f"_{counter:02}"
@@ -529,20 +515,13 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
     def delete(
         self,
         *args,
-        transaction_id=None,
         **kwargs,
     ):
-        # Versions must share the root's transaction ID so they are restored
+        # If deleting a root document, move all its versions to trash as well.
-        # together by django-softdelete.
-        if transaction_id is None:
-            transaction_id = uuid.uuid4()
         if self.root_document_id is None:
-            Document.objects.filter(root_document=self).delete(
+            Document.objects.filter(root_document=self).delete()
-                transaction_id=transaction_id,
-            )
         return super().delete(
             *args,
-            transaction_id=transaction_id,
             **kwargs,
         )
 

src/documents/permissions.py

@@ -173,179 +173,6 @@ def set_permissions_for_object(
                         )
 
 
-def _resolve_permissions(codenames: set[str], ctype: ContentType) -> list[Permission]:
-    """
-    Resolves `codenames` to Permission rows, raising like the single-object
-    assign_perm() this bulk path replaces does (via a `.get()` internally)
-    if any codename doesn't exist -- e.g. a client-supplied action name that
-    was never validated (BulkEditObjectsSerializer._validate_permissions
-    calls validate_set_permissions() only for its side-effecting id checks
-    and discards the filtered dict it returns, so an unrecognized action key
-    reaches this function as-is). A plain `.filter()` with no existence
-    check would otherwise silently build zero rows and no-op instead of
-    reporting the bad input.
-    """
-    permission_objs = list(
-        Permission.objects.filter(content_type=ctype, codename__in=codenames),
-    )
-    missing = codenames - {p.codename for p in permission_objs}
-    if missing:
-        raise Permission.DoesNotExist(
-            f"Permission matching query does not exist for codename(s): "
-            f"{', '.join(sorted(missing))}",
-        )
-    return permission_objs
-
-
-def _apply_bulk_permission_entry(
-    *,
-    perm_model: type[UserObjectPermission] | type[GroupObjectPermission],
-    identity_model: type[User] | type[Group],
-    identity_field: str,
-    ids: list[int],
-    codename: str,
-    permission_objs: list[Permission],
-    ctype: ContentType,
-    object_pks: list[str],
-    merge: bool,
-) -> None:
-    # Only the ids are needed to build permission rows (via `<field>_id=`),
-    # so avoid fetching full User/Group rows for identities that may not
-    # even end up being granted anything new.
-    add_ids = set(
-        identity_model.objects.filter(id__in=ids).values_list("id", flat=True),
-    )
-
-    if not merge:
-        existing_ids = set(
-            perm_model.objects.filter(
-                content_type=ctype,
-                object_pk__in=object_pks,
-                permission__codename=codename,
-            )
-            .values_list(f"{identity_field}_id", flat=True)
-            .distinct(),
-        )
-        remove_ids = existing_ids - add_ids
-        if remove_ids:
-            perm_model.objects.filter(
-                content_type=ctype,
-                object_pk__in=object_pks,
-                permission__codename=codename,
-                **{f"{identity_field}_id__in": remove_ids},
-            ).delete()
-
-    if not add_ids:
-        return
-
-    rows = [
-        perm_model(
-            content_type=ctype,
-            object_pk=pk,
-            permission=permission_obj,
-            **{f"{identity_field}_id": identity_id},
-        )
-        for permission_obj in permission_objs
-        for pk in object_pks
-        for identity_id in add_ids
-    ]
-    # ignore_conflicts skips only rows that already exist as an exact
-    # (identity, permission, object) match -- the same de-dup the
-    # underlying (user|group, permission, object_pk) unique constraint
-    # already enforces for the single-object assign_perm() this replaces,
-    # so it doesn't change what counts as "already granted". batch_size
-    # caps how many rows go into a single INSERT statement.
-    perm_model.objects.bulk_create(rows, ignore_conflicts=True, batch_size=1000)
-
-
-def set_permissions_for_objects(
-    permissions: dict,
-    model: type[Model],
-    pks: QuerySet | list,
-    *,
-    merge: bool = False,
-) -> None:
-    """
-    Bulk equivalent of set_permissions_for_object: applies the same
-    permission changes to every object identified by `pks` at once.
-
-    Takes a model + pks (rather than model instances) deliberately -- the
-    permission rows built below only ever need `pk`, `content_type`, and
-    identity ids, so callers shouldn't have to fetch full rows (with every
-    other field) just to hand them to this function.
-
-    Deliberately does not use guardian's queryset/list-aware assign_perm:
-    passing a list as the object routes to bulk_assign_perm, which skips
-    creating a direct permission row for anyone who already has the
-    permission via ANY group membership (it checks
-    ObjectPermissionChecker.has_perm, which is group-inheritance-aware) --
-    unlike the single-object assign_perm this replaces, which always
-    ensures a direct row via get_or_create regardless of group-derived
-    access. Losing that guarantee would mean a later revocation of the
-    group's grant silently strips access an admin explicitly asked to be
-    direct. Bulk-creating rows straight against the permission models
-    instead (see _apply_bulk_permission_entry) preserves the original
-    always-create-a-direct-row semantics while still batching every object
-    and every identity into one query per action, rather than one query per
-    (object, user) pair.
-    """
-    object_pks = [str(pk) for pk in pks]
-    if not object_pks:  # pragma: no cover
-        return
-
-    model_name = model.__name__.lower()
-    ctype = ContentType.objects.get_for_model(model)
-
-    # Every action is resolved up front, before anything is written, so an
-    # unrecognized action name (see _resolve_permissions) aborts the whole
-    # call instead of leaving the actions ahead of it already applied --
-    # BulkEditObjectsSerializer lets unknown keys through and its view turns
-    # the exception into a 400, so a half-applied change would otherwise be
-    # reported to the client as a failure.
-    permissions_by_action: dict[str, list[Permission]] = {}
-    for action, entry in permissions.items():
-        if "users" not in entry and "groups" not in entry:
-            continue
-        implied_codenames = {f"{action}_{model_name}"}
-        if action == "change":
-            # change gives view too
-            implied_codenames.add(f"view_{model_name}")
-        permissions_by_action[action] = _resolve_permissions(
-            implied_codenames,
-            ctype,
-        )
-
-    for action, entry in permissions.items():
-        codename = f"{action}_{model_name}"
-        permission_objs = permissions_by_action.get(action, [])
-
-        if "users" in entry:
-            _apply_bulk_permission_entry(
-                perm_model=UserObjectPermission,
-                identity_model=User,
-                identity_field="user",
-                ids=entry["users"],
-                codename=codename,
-                permission_objs=permission_objs,
-                ctype=ctype,
-                object_pks=object_pks,
-                merge=merge,
-            )
-
-        if "groups" in entry:
-            _apply_bulk_permission_entry(
-                perm_model=GroupObjectPermission,
-                identity_model=Group,
-                identity_field="group",
-                ids=entry["groups"],
-                codename=codename,
-                permission_objs=permission_objs,
-                ctype=ctype,
-                object_pks=object_pks,
-                merge=merge,
-            )
-
-
 def permitted_object_ids(
     user: User | None,
     model: type[Model],

src/documents/serialisers.py

@@ -3,7 +3,6 @@ from __future__ import annotations
 import logging
 import math
 import re
-from collections.abc import Iterable
 from datetime import datetime
 from datetime import timedelta
 from decimal import Decimal
@@ -88,9 +87,9 @@ from documents.regex import validate_regex_pattern
 from documents.templating.filepath import validate_filepath_template_and_render
 from documents.templating.utils import convert_format_str_to_template_format
 from documents.templating.workflows import validate_workflow_template
+from documents.utils import normalize_unicode
 from documents.validators import uri_validator
 from documents.validators import url_validator
-from documents.versioning import has_prefetched_effective_content
 from documents.versioning import sort_versions_newest_first
 
 if TYPE_CHECKING:
@@ -884,86 +883,8 @@ def validate_documentlink_targets(user, doc_ids):
         )
 
 
-# drf-writable-nested revalidates a document's custom_fields more than once
-# per request: once as the ordinary nested list, then again per-item while
-# matching existing vs. new CustomFieldInstance rows during save() -- and
-# that second pass builds a brand new serializer (and field) instance per
-# item (see its update_or_create_reverse_relations / _get_serializer_for_field),
-# so a cache on the field instance alone only helps the first pass. It does,
-# however, explicitly pass `context=self.context` to every one of those
-# fresh serializers -- the *same* dict object the outer DocumentSerializer
-# is using, not a copy. That context dict is already request-scoped (DRF
-# builds it fresh per request via get_serializer_context()), so stashing the
-# resolved CustomField objects there -- rather than in some new global/
-# thread-local cache -- lets every later pass reuse them for free while
-# staying entirely within DRF's existing, already-request-scoped machinery.
-_CUSTOM_FIELD_CONTEXT_CACHE_KEY = "_custom_field_lookup_cache"
-
-
-class _CachingCustomFieldPrimaryKeyField(serializers.PrimaryKeyRelatedField):
-    """
-    Resolves CustomField ids via a cache on the serializer context, so
-    later, separately-instantiated fields for the same request (drf-writable-
-    nested rebuilds one per item during save) reuse what was already
-    resolved instead of re-querying.
-    """
-
-    def _cache(self) -> dict[int, CustomField]:
-        return self.context.setdefault(_CUSTOM_FIELD_CONTEXT_CACHE_KEY, {})
-
-    @staticmethod
-    def _normalize_pk(data: Any) -> int | None:
-        """
-        Returns `data` coerced to the int a valid CustomField pk would be,
-        or None if `data` isn't a plausible pk (wrong type, unhashable,
-        non-numeric, or a bool -- DRF itself rejects bools as pks since
-        `True == 1` would otherwise silently match). None tells callers to
-        leave `data` alone and let `super().to_internal_value()` report the
-        normal validation error instead of touching the cache/queryset with
-        it directly.
-        """
-        if isinstance(data, bool):
-            return None
-        try:
-            return int(data)
-        except (TypeError, ValueError):
-            return None
-
-    def prefetch(self, ids: Iterable[Any]) -> None:
-        cache = self._cache()
-        candidates = {pk for i in ids if (pk := self._normalize_pk(i)) is not None}
-        missing = {i for i in candidates if i not in cache}
-        if missing:
-            for obj in self.get_queryset().filter(pk__in=missing):
-                cache[obj.pk] = obj
-
-    def to_internal_value(self, data: Any) -> CustomField:
-        pk = self._normalize_pk(data)
-        if pk is None:
-            return super().to_internal_value(data)
-        cache = self._cache()
-        if pk in cache:
-            return cache[pk]
-        obj: CustomField = super().to_internal_value(data)
-        cache[obj.pk] = obj
-        return obj
-
-
-class CustomFieldInstanceListSerializer(serializers.ListSerializer):
-    def to_internal_value(self, data: Any) -> list[Any]:
-        if isinstance(data, list):
-            field_ids = [
-                item["field"]
-                for item in data
-                if isinstance(item, dict) and "field" in item
-            ]
-            if field_ids:
-                self.child.fields["field"].prefetch(field_ids)
-        return super().to_internal_value(data)
-
-
 class CustomFieldInstanceSerializer(serializers.ModelSerializer[CustomFieldInstance]):
-    field = _CachingCustomFieldPrimaryKeyField(queryset=CustomField.objects.all())
+    field = serializers.PrimaryKeyRelatedField(queryset=CustomField.objects.all())
     value = ReadWriteSerializerMethodField(allow_null=True)
 
     def create(self, validated_data):
@@ -1064,7 +985,6 @@ class CustomFieldInstanceSerializer(serializers.ModelSerializer[CustomFieldInsta
 
     class Meta:
         model = CustomFieldInstance
-        list_serializer_class = CustomFieldInstanceListSerializer
         fields = [
             "value",
             "field",
@@ -1233,14 +1153,8 @@ class DocumentSerializer(
 
     def to_representation(self, instance):
         doc = super().to_representation(instance)
-        if "content" in self.fields and has_prefetched_effective_content(instance):
+        if "content" in self.fields and hasattr(instance, "effective_content"):
-            # Only resolve version-aware content when it's cheap: an SQL
+            doc["content"] = getattr(instance, "effective_content") or ""
-            # annotation or a versions prefetch is already on the instance.
-            # A caller that set up neither (e.g. TrashView, GlobalSearchView,
-            # which build their own querysets) gets the document's own,
-            # unresolved content instead of paying for an extra per-instance
-            # query -- same as before effective_content resolution existed.
-            doc["content"] = instance.get_effective_content() or ""
         if self.truncate_content and "content" in self.fields:
             doc["content"] = doc.get("content")[0:550]
         return doc
@@ -1334,31 +1248,30 @@ class DocumentSerializer(
 
             validated_data["tags"] = list(final_tags)
         if validated_data.get("remove_inbox_tags"):
-            current_tag_ids = {t.pk for t in instance.tags.all()}
+            tag_ids_being_added = (
-            tags = (
+                [
-                validated_data["tags"]
+                    tag.id
+                    for tag in validated_data["tags"]
+                    if tag not in instance.tags.all()
+                ]
                 if "tags" in validated_data
-                else list(instance.tags.all())
+                else []
             )
-
+            inbox_tags_not_being_added = Tag.objects.filter(is_inbox_tag=True).exclude(
-            # Tags newly added in this update, plus their ancestors, are kept
+                id__in=tag_ids_being_added,
-            keep_ids: set[int] = set()
+            )
-            for tag in tags:
+            if "tags" in validated_data:
-                if tag.pk not in current_tag_ids:
+                validated_data["tags"] = [
-                    keep_ids.add(tag.pk)
+                    tag
-                    keep_ids.update(int(pk) for pk in tag.get_ancestors_pks())
+                    for tag in validated_data["tags"]
-
+                    if tag not in inbox_tags_not_being_added
-            # Remove inbox tags and their descendants, except those being kept
+                ]
-            remove_ids: set[int] = set()
+            else:
-            for inbox_tag in (
+                validated_data["tags"] = [
-                Tag.objects.filter(is_inbox_tag=True)
+                    tag
-                .exclude(pk__in=keep_ids)
+                    for tag in instance.tags.all()
-                .only("pk", "tn_descendants_pks")
+                    if tag not in inbox_tags_not_being_added
-            ):
+                ]
-                remove_ids.add(inbox_tag.pk)
-                remove_ids.update(int(pk) for pk in inbox_tag.get_descendants_pks())
-
-            validated_data["tags"] = [t for t in tags if t.pk not in remove_ids]
 
         if settings.AUDIT_LOG_ENABLED:
             with set_actor(self.user):
@@ -3208,6 +3121,13 @@ class WorkflowTriggerSerializer(serializers.ModelSerializer[WorkflowTrigger]):
         ):
             attrs["filter_path"] = None
 
+        # Normalize once at write time, since these are matched against many
+        # documents but edited rarely
+        if attrs.get("filter_filename") is not None:
+            attrs["filter_filename"] = normalize_unicode(attrs["filter_filename"])
+        if attrs.get("filter_path") is not None:
+            attrs["filter_path"] = normalize_unicode(attrs["filter_path"])
+
         if (
             "filter_custom_field_query" in attrs
             and attrs["filter_custom_field_query"] is not None

src/documents/templating/filepath.py

@@ -1,7 +1,6 @@
 import logging
 import os
 import re
-import unicodedata
 from collections.abc import Iterable
 from pathlib import PurePath
 
@@ -26,6 +25,7 @@ from documents.templating.environment import _template_environment
 from documents.templating.filters import format_datetime
 from documents.templating.filters import get_cf_value
 from documents.templating.filters import localize_date
+from documents.utils import normalize_unicode
 
 logger = logging.getLogger("paperless.templating")
 
@@ -42,7 +42,7 @@ class FilePathTemplate(Template):
             3. Removing extra spaces before and after forward slashes
             4. Preserving spaces in other parts of the path
             """
-            value = unicodedata.normalize("NFC", value)
+            value = normalize_unicode(value)
             value = value.replace("\n", "").replace("\r", "")
             value = re.sub(r"\s*/\s*", "/", value)
 
@@ -184,17 +184,17 @@ def get_basic_metadata_context(
     """
     return {
         "title": pathvalidate.sanitize_filename(
-            unicodedata.normalize("NFC", document.title),
+            normalize_unicode(document.title),
             replacement_text="-",
         ),
         "correspondent": pathvalidate.sanitize_filename(
-            unicodedata.normalize("NFC", document.correspondent.name),
+            normalize_unicode(document.correspondent.name),
             replacement_text="-",
         )
         if document.correspondent
         else no_value_default,
         "document_type": pathvalidate.sanitize_filename(
-            unicodedata.normalize("NFC", document.document_type.name),
+            normalize_unicode(document.document_type.name),
             replacement_text="-",
         )
         if document.document_type
@@ -205,8 +205,7 @@ def get_basic_metadata_context(
         "owner_username": document.owner.username
         if document.owner
         else no_value_default,
-        "original_name": unicodedata.normalize(
+        "original_name": normalize_unicode(
-            "NFC",
             PurePath(document.original_filename).with_suffix("").name,
         )
         if document.original_filename
@@ -275,12 +274,12 @@ def get_tags_context(tags: Iterable[Tag]) -> dict[str, str | list[str]]:
     return {
         "tag_list": pathvalidate.sanitize_filename(
             ",".join(
-                sorted(unicodedata.normalize("NFC", tag.name) for tag in tags),
+                sorted(normalize_unicode(tag.name) for tag in tags),
             ),
             replacement_text="-",
         ),
         # Assumed to be ordered, but a template could loop through to find what they want
-        "tag_name_list": [unicodedata.normalize("NFC", x.name) for x in tags],
+        "tag_name_list": [normalize_unicode(x.name) for x in tags],
     }
 
 
@@ -307,7 +306,7 @@ def get_custom_fields_context(
             CustomField.FieldDataType.LONG_TEXT,
         }:
             value = pathvalidate.sanitize_filename(
-                unicodedata.normalize("NFC", field_instance.value),
+                normalize_unicode(field_instance.value),
                 replacement_text="-",
             )
         elif (
@@ -316,8 +315,7 @@ def get_custom_fields_context(
         ):
             options = field_instance.field.extra_data["select_options"]
             value = pathvalidate.sanitize_filename(
-                unicodedata.normalize(
+                normalize_unicode(
-                    "NFC",
                     next(
                         option["label"]
                         for option in options
@@ -330,7 +328,7 @@ def get_custom_fields_context(
             value = field_instance.value
         field_data["custom_fields"][
             pathvalidate.sanitize_filename(
-                unicodedata.normalize("NFC", field_instance.field.name),
+                normalize_unicode(field_instance.field.name),
                 replacement_text="-",
             )
         ] = {

src/documents/tests/test_api_custom_fields.py

@@ -5,19 +5,14 @@ from unittest.mock import ANY
 
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from django.db import connection
 from django.test import override_settings
-from django.test.utils import CaptureQueriesContext
 from guardian.shortcuts import assign_perm
 from rest_framework import status
 from rest_framework.test import APITestCase
 
-from documents.data_models import DocumentMetadataOverrides
 from documents.models import CustomField
 from documents.models import CustomFieldInstance
 from documents.models import Document
-from documents.serialisers import DocumentSerializer
-from documents.tests.factories import DocumentFactory
 from documents.tests.utils import DirectoriesMixin
 
 
@@ -535,190 +530,6 @@ class TestCustomFieldsAPI(DirectoriesMixin, APITestCase):
         doc.refresh_from_db()
         self.assertEqual(len(doc.custom_fields.all()), 10)
 
-    def test_document_serializer_custom_fields_validation_batches_field_lookup(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A document is being validated with several custom field values
-              at once (as happens on every PATCH/PUT/POST)
-        WHEN:
-            - The serializer is validated
-        THEN:
-            - The referenced CustomField objects are resolved with a single
-              query, not one query per custom field
-        """
-        doc = DocumentFactory(mime_type="application/pdf")
-        custom_fields = [
-            CustomField.objects.create(
-                name=f"Test Custom Field {i}",
-                data_type=CustomField.FieldDataType.STRING,
-            )
-            for i in range(5)
-        ]
-
-        serializer = DocumentSerializer(
-            doc,
-            data={
-                "custom_fields": [
-                    {"field": custom_field.id, "value": "test value"}
-                    for custom_field in custom_fields
-                ],
-            },
-            partial=True,
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            self.assertTrue(serializer.is_valid(), serializer.errors)
-
-        custom_field_lookups = [
-            query
-            for query in ctx.captured_queries
-            if 'FROM "documents_customfield" WHERE "documents_customfield"."id"'
-            in query["sql"]
-        ]
-        self.assertEqual(
-            len(custom_field_lookups),
-            1,
-            "Expected a single batched query to resolve the custom fields, "
-            f"got {len(custom_field_lookups)}: {custom_field_lookups}",
-        )
-
-    def test_document_serializer_save_reuses_cached_custom_fields(self) -> None:
-        """
-        GIVEN:
-            - A document is being saved with several custom field values via
-              DocumentSerializer, which drives drf-writable-nested's real
-              update_or_create_reverse_relations path -- rebuilding a fresh
-              CustomFieldInstanceSerializer per item during save(), the exact
-              mechanism the shared-context cache exists to optimize
-        WHEN:
-            - The serializer, already validated, is saved
-        THEN:
-            - No further CustomField queries are issued: each per-item
-              nested serializer reuses the CustomField objects resolved
-              during is_valid(), instead of re-resolving them during save()
-        """
-        doc = DocumentFactory(mime_type="application/pdf")
-        custom_fields = [
-            CustomField.objects.create(
-                name=f"Test Custom Field {i}",
-                data_type=CustomField.FieldDataType.STRING,
-            )
-            for i in range(5)
-        ]
-
-        serializer = DocumentSerializer(
-            doc,
-            data={
-                "custom_fields": [
-                    {"field": custom_field.id, "value": "test value"}
-                    for custom_field in custom_fields
-                ],
-            },
-            partial=True,
-        )
-        self.assertTrue(serializer.is_valid(), serializer.errors)
-
-        with CaptureQueriesContext(connection) as ctx:
-            serializer.save()
-
-        custom_field_lookups = [
-            query
-            for query in ctx.captured_queries
-            if 'FROM "documents_customfield" WHERE "documents_customfield"."id"'
-            in query["sql"]
-        ]
-        self.assertEqual(
-            custom_field_lookups,
-            [],
-            "Expected save() to reuse CustomField objects resolved during "
-            f"is_valid(), got: {custom_field_lookups}",
-        )
-        self.assertEqual(doc.custom_fields.count(), 5)
-
-    def test_custom_field_validation_rejects_malformed_field_value(self) -> None:
-        """
-        GIVEN:
-            - A document is being validated with a malformed custom_fields
-              entry whose "field" value is neither a valid CustomField id
-              nor a type DRF's own PrimaryKeyRelatedField can safely reject
-              on its own (unhashable, or a non-numeric scalar)
-        WHEN:
-            - The serializer is validated
-        THEN:
-            - A normal validation error is raised, not an unhandled
-              TypeError/ValueError escaping past DRF's validation layer
-        """
-        doc = DocumentFactory(mime_type="application/pdf")
-
-        bad_field_values = {
-            "unhashable-list": [],
-            "unhashable-dict": {},
-            "non-numeric-scalar": "abc",
-        }
-        for case_id, bad_field_value in bad_field_values.items():
-            with self.subTest(case_id):
-                serializer = DocumentSerializer(
-                    doc,
-                    data={
-                        "custom_fields": [
-                            {"field": bad_field_value, "value": "test value"},
-                        ],
-                    },
-                    partial=True,
-                )
-
-                self.assertFalse(serializer.is_valid())
-                self.assertIn("custom_fields", serializer.errors)
-
-    def test_document_metadata_overrides_from_document_batches_field_lookup(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A document has several custom field values
-            - The document instance has just been refreshed from the database,
-              which drops any prefetched related objects (as
-              send_websocket_document_updated does before building overrides)
-        WHEN:
-            - DocumentMetadataOverrides.from_document() reads the document's
-              custom field values
-        THEN:
-            - The referenced CustomField objects are resolved with a single
-              query, not one query per custom field
-        """
-        doc = DocumentFactory(mime_type="application/pdf")
-        for i in range(5):
-            CustomFieldInstance.objects.create(
-                document=doc,
-                field=CustomField.objects.create(
-                    name=f"Test Custom Field {i}",
-                    data_type=CustomField.FieldDataType.STRING,
-                ),
-                value_text="value",
-            )
-
-        doc.refresh_from_db()
-
-        with CaptureQueriesContext(connection) as ctx:
-            overrides = DocumentMetadataOverrides.from_document(doc)
-
-        self.assertEqual(len(overrides.custom_fields), 5)
-        unbatched_field_lookups = [
-            query
-            for query in ctx.captured_queries
-            if 'FROM "documents_customfield" WHERE "documents_customfield"."id"'
-            in query["sql"]
-        ]
-        self.assertEqual(
-            unbatched_field_lookups,
-            [],
-            "Expected CustomField data to come from the CustomFieldInstance "
-            "join, not a separate per-instance lookup, "
-            f"got: {unbatched_field_lookups}",
-        )
-
     def test_change_custom_field_instance_value(self) -> None:
         """
         GIVEN:

src/documents/tests/test_api_objects.py

@@ -2,15 +2,10 @@ import datetime
 import json
 from unittest import mock
 
-from django.contrib.auth.models import Group
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from django.db import connection
 from django.test import override_settings
-from django.test.utils import CaptureQueriesContext
 from guardian.shortcuts import assign_perm
-from guardian.shortcuts import get_groups_with_perms
-from guardian.shortcuts import get_users_with_perms
 from rest_framework import status
 from rest_framework.test import APITestCase
 
@@ -847,66 +842,6 @@ class TestBulkEditObjects(APITestCase):
         self.assertEqual(response.status_code, status.HTTP_200_OK)
         self.assertEqual(StoragePath.objects.count(), 0)
 
-    def test_bulk_objects_set_permissions_batched_across_object_count(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - Many tags are being bulk-edited to set permissions at once
-        WHEN:
-            - bulk_edit_objects API endpoint is called with set_permissions
-              operation over a small batch vs. a much larger one
-        THEN:
-            - Permissions are applied correctly at both scales
-            - Query count does not grow with the number of tags, i.e. each
-              user/group is applied across all tags with one batched call
-              rather than one call per (tag, identity) pair
-        """
-        group1 = Group.objects.create(name="perm-group")
-        permissions = {
-            "view": {"users": [self.user1.id, self.user2.id], "groups": [group1.id]},
-            "change": {"users": [self.user1.id], "groups": [group1.id]},
-        }
-
-        def run_with_n_tags(n: int) -> int:
-            tags = [Tag.objects.create(name=f"perm-tag-{n}-{i}") for i in range(n)]
-            with CaptureQueriesContext(connection) as ctx:
-                response = self.client.post(
-                    "/api/bulk_edit_objects/",
-                    json.dumps(
-                        {
-                            "objects": [t.id for t in tags],
-                            "object_type": "tags",
-                            "operation": "set_permissions",
-                            "permissions": permissions,
-                            "merge": False,
-                        },
-                    ),
-                    content_type="application/json",
-                )
-            self.assertEqual(response.status_code, status.HTTP_200_OK)
-            for tag in tags:
-                self.assertEqual(get_users_with_perms(tag).count(), 2)
-                self.assertEqual(get_groups_with_perms(tag).count(), 1)
-            return len(ctx.captured_queries)
-
-        small_batch_queries = run_with_n_tags(5)
-        large_batch_queries = run_with_n_tags(50)
-
-        # A tolerance rather than equality, matching the N+1 check in
-        # test_views.py: bulk_create's batch_size caps rows per INSERT, so a
-        # large enough selection does legitimately add statements, and the
-        # per-process ContentType cache makes the first run carry an extra
-        # query. Neither can hide a regression to per-object assignment,
-        # which would be ~10x the small-batch count here.
-        self.assertLessEqual(
-            large_batch_queries,
-            small_batch_queries + 5,
-            "Permission assignment appears to scale with object count: "
-            f"{small_batch_queries} queries for 5 tags vs. "
-            f"{large_batch_queries} for 50",
-        )
-
     def test_bulk_objects_delete_all_filtered(self) -> None:
         """
         GIVEN:

src/documents/tests/test_api_trash.py

@@ -207,65 +207,3 @@ class TestTrashAPI(DirectoriesMixin, APITestCase):
         )
         self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
         self.assertIn("have not yet been deleted", resp.data["documents"][0])
-
-    def _make_versioned_document(self) -> tuple[Document, list[Document]]:
-        root = Document.objects.create(
-            title="root",
-            content="root-content",
-            checksum="root",
-            mime_type="application/pdf",
-        )
-        versions = [
-            Document.objects.create(
-                title=f"v{index}",
-                content=f"v{index}-content",
-                checksum=f"v{index}",
-                mime_type="application/pdf",
-                root_document=root,
-                version_index=index,
-            )
-            for index in range(1, 3)
-        ]
-        return root, versions
-
-    def test_api_trash_restore_document_restores_its_versions(self) -> None:
-        """
-        GIVEN:
-            - Existing document with two versions
-        WHEN:
-            - API request to delete the document
-            - API request to restore it from the trash
-        THEN:
-            - Only the document itself is listed in the trash
-            - A version cannot be restored without its root
-            - The document is restored together with all of its versions
-        """
-        root, versions = self._make_versioned_document()
-
-        self.client.force_login(user=self.user)
-        self.client.delete(f"/api/documents/{root.pk}/")
-        self.assertEqual(Document.deleted_objects.count(), 3)
-
-        resp = self.client.get("/api/trash/")
-        self.assertEqual(resp.status_code, status.HTTP_200_OK)
-        self.assertEqual(resp.data["count"], 1)
-        self.assertEqual(resp.data["results"][0]["id"], root.pk)
-
-        # A version cannot be restored while its root remains in the trash.
-        resp = self.client.post(
-            "/api/trash/",
-            {"action": "restore", "documents": [versions[0].pk]},
-        )
-        self.assertEqual(resp.status_code, status.HTTP_400_BAD_REQUEST)
-        self.assertIn("Restore the root document", resp.data["documents"][0])
-
-        resp = self.client.post(
-            "/api/trash/",
-            {"action": "restore", "documents": [root.pk]},
-        )
-        self.assertEqual(resp.status_code, status.HTTP_200_OK)
-        self.assertEqual(Document.deleted_objects.count(), 0)
-        self.assertCountEqual(
-            Document.objects.filter(root_document=root).values_list("id", flat=True),
-            [version.pk for version in versions],
-        )

src/documents/tests/test_api_update_version_nfc.py

@@ -0,0 +1,69 @@
+import unicodedata
+from typing import TYPE_CHECKING
+from unittest import mock
+
+import celery.result
+import pytest
+from django.core.files.uploadedfile import SimpleUploadedFile
+
+from documents.models import Document
+
+if TYPE_CHECKING:
+    from documents.data_models import ConsumableDocument
+
+
+@pytest.fixture()
+def consume_file_mock():
+    with mock.patch("documents.tasks.consume_file.apply_async") as m:
+        m.return_value = celery.result.AsyncResult(id="test-task-id")
+        yield m
+
+
+@pytest.fixture()
+def directories(tmp_path, settings, _media_settings):
+    scratch = tmp_path / "scratch"
+    scratch.mkdir()
+    settings.SCRATCH_DIR = scratch
+    return scratch
+
+
+@pytest.mark.django_db
+class TestUpdateVersionNFCNormalization:
+    def test_nfd_filename_normalized_to_nfc(
+        self,
+        admin_client,
+        consume_file_mock: mock.MagicMock,
+        directories,
+    ):
+        """Uploaded new-version file with NFD filename must have its temp name stored as NFC."""
+        document = Document.objects.create(
+            title="Test",
+            content="content",
+            checksum="checksum",
+            mime_type="application/pdf",
+        )
+
+        nfd = unicodedata.normalize("NFD", "Rechnung März.pdf")
+        nfc = unicodedata.normalize("NFC", "Rechnung März.pdf")
+
+        assert nfd != nfc
+
+        uploaded = SimpleUploadedFile(
+            nfd,
+            b"%PDF-1.4 test",
+            content_type="application/pdf",
+        )
+        response = admin_client.post(
+            f"/api/documents/{document.pk}/update_version/",
+            {"document": uploaded},
+        )
+
+        assert response.status_code == 200
+
+        task_kwargs = consume_file_mock.call_args.kwargs["kwargs"]
+        input_doc: ConsumableDocument = task_kwargs["input_doc"]
+
+        assert input_doc.original_file.name == nfc, (
+            f"Expected NFC filename {nfc!r}, got {input_doc.original_file.name!r}"
+        )
+        assert unicodedata.is_normalized("NFC", input_doc.original_file.name)

src/documents/tests/test_bulk_edit.py

@@ -5,11 +5,8 @@ from unittest import mock
 
 import pikepdf
 from django.contrib.auth.models import Group
-from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from django.db import connection
 from django.test import TestCase
-from django.test.utils import CaptureQueriesContext
 from guardian.shortcuts import assign_perm
 from guardian.shortcuts import get_groups_with_perms
 from guardian.shortcuts import get_users_with_perms
@@ -22,7 +19,6 @@ from documents.models import Document
 from documents.models import DocumentType
 from documents.models import StoragePath
 from documents.models import Tag
-from documents.permissions import set_permissions_for_objects
 from documents.tests.utils import DirectoriesMixin
 
 
@@ -396,11 +392,6 @@ class TestBulkEdit(DirectoriesMixin, TestCase):
         self.assertFalse(Document.objects.filter(id=self.doc1.id).exists())
         self.assertFalse(Document.objects.filter(id=version.id).exists())
 
-        Document.deleted_objects.get(id=self.doc1.id).restore(strict=False)
-
-        self.assertTrue(Document.objects.filter(id=self.doc1.id).exists())
-        self.assertTrue(Document.objects.filter(id=version.id).exists())
-
     def test_delete_version_document_keeps_root(self) -> None:
         version = Document.objects.create(
             checksum="A-v1",
@@ -519,178 +510,6 @@ class TestBulkEdit(DirectoriesMixin, TestCase):
         )
         self.assertEqual(groups_with_perms.count(), 2)
 
-    @mock.patch("documents.tasks.bulk_update_documents.apply_async")
-    def test_set_permissions_batched_across_document_count(
-        self,
-        m,
-    ) -> None:
-        """
-        GIVEN:
-            - Many documents are being bulk-edited to set permissions at once
-        WHEN:
-            - set_permissions runs over a small batch vs. a much larger one
-        THEN:
-            - Permissions are applied correctly at both scales
-            - Query count does not grow with the number of documents, i.e.
-              each user/group is applied across all documents with one
-              batched call rather than one call per (document, identity)
-              pair
-        """
-        permissions = {
-            "view": {
-                "users": [self.user1.id, self.user2.id],
-                "groups": [self.group2.id],
-            },
-            "change": {
-                "users": [self.user1.id],
-                "groups": [self.group2.id],
-            },
-        }
-
-        def run_with_n_documents(n: int) -> int:
-            docs = [
-                Document.objects.create(checksum=f"perm-{n}-{i}", title=f"perm-{n}-{i}")
-                for i in range(n)
-            ]
-            with CaptureQueriesContext(connection) as ctx:
-                bulk_edit.set_permissions(
-                    [doc.id for doc in docs],
-                    set_permissions=permissions,
-                    owner=self.owner,
-                    merge=False,
-                )
-            for doc in docs:
-                self.assertEqual(get_users_with_perms(doc).count(), 2)
-                self.assertEqual(get_groups_with_perms(doc).count(), 1)
-            return len(ctx.captured_queries)
-
-        small_batch_queries = run_with_n_documents(5)
-        large_batch_queries = run_with_n_documents(50)
-
-        # A tolerance rather than equality, matching the N+1 check in
-        # test_views.py: bulk_create's batch_size caps rows per INSERT, so a
-        # large enough selection does legitimately add statements, and the
-        # per-process ContentType cache makes the first run carry an extra
-        # query. Neither can hide a regression to per-document assignment,
-        # which would be ~10x the small-batch count here.
-        self.assertLessEqual(
-            large_batch_queries,
-            small_batch_queries + 5,
-            "Permission assignment appears to scale with document count: "
-            f"{small_batch_queries} queries for 5 documents vs. "
-            f"{large_batch_queries} for 50",
-        )
-
-    @mock.patch("documents.tasks.bulk_update_documents.apply_async")
-    def test_set_permissions_grants_direct_perm_even_if_already_granted_via_group(
-        self,
-        m,
-    ) -> None:
-        """
-        GIVEN:
-            - A user already has view access to a document via group
-              membership, with no direct grant of their own
-        WHEN:
-            - set_permissions explicitly grants that same user direct view
-              access via bulk_edit
-        THEN:
-            - A direct permission grant is created for the user, not skipped
-              because they already have equivalent access via the group
-
-        Regression test: guardian's queryset-aware assign_perm() (routed to
-        when the target is a list/queryset) skips creating a direct row for
-        anyone whose ObjectPermissionChecker.has_perm() already returns True
-        -- which includes group-derived access. The single-object assign_perm
-        this bulk path replaces has no such check; it always ensures a
-        direct row via get_or_create. Losing that guarantee would mean
-        revoking the group's grant later silently strips access that was
-        supposed to be explicit.
-        """
-        self.doc1.owner = self.user1
-        self.doc1.save()
-        self.user1.groups.add(self.group1)
-        assign_perm("view_document", self.group1, self.doc1)
-
-        bulk_edit.set_permissions(
-            [self.doc1.id],
-            set_permissions={
-                "view": {"users": [self.user1.id], "groups": []},
-            },
-            merge=True,
-        )
-
-        direct_users = get_users_with_perms(
-            self.doc1,
-            only_with_perms_in=["view_document"],
-            with_group_users=False,
-        )
-        self.assertIn(self.user1, direct_users)
-
-    def test_set_permissions_for_objects_raises_for_unknown_action(self) -> None:
-        """
-        GIVEN:
-            - An unrecognized permission action name with users to grant it
-              to
-        WHEN:
-            - set_permissions_for_objects is called
-        THEN:
-            - Permission.DoesNotExist is raised, not a silent no-op
-
-        Regression test: the endpoint that calls this
-        (BulkEditObjectPermissionsView) never actually validates action
-        names against the raw client-supplied permissions dict --
-        BulkEditObjectsSerializer._validate_permissions calls
-        validate_set_permissions() only for its side-effecting user/group id
-        checks and discards the filtered dict it returns -- so a bogus
-        action key reaches this function as-is. Resolving the Permission via
-        a bare `.filter()` (which returns empty instead of raising) would
-        silently drop the grant and report success.
-        """
-        with self.assertRaises(Permission.DoesNotExist):
-            set_permissions_for_objects(
-                {"not_a_real_action": {"users": [self.user1.id], "groups": []}},
-                Document,
-                [self.doc1.pk],
-            )
-
-    def test_set_permissions_for_objects_unknown_action_applies_nothing(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A permissions dict with a valid action ordered ahead of an
-              unrecognized one
-        WHEN:
-            - set_permissions_for_objects is called
-        THEN:
-            - Permission.DoesNotExist is raised
-            - The valid action ahead of it is not applied either
-
-        Every action is resolved before any row is written, so a bad action
-        name cannot leave a half-applied change behind. That matters because
-        BulkEditObjectsView turns this exception into a 400: without the
-        up-front resolution the client would be told the request failed
-        while the leading action had already been committed.
-        """
-        with self.assertRaises(Permission.DoesNotExist):
-            set_permissions_for_objects(
-                {
-                    "view": {"users": [self.user1.id], "groups": []},
-                    "not_a_real_action": {"users": [self.user1.id], "groups": []},
-                },
-                Document,
-                [self.doc1.pk],
-            )
-
-        self.assertNotIn(
-            self.user1,
-            get_users_with_perms(
-                self.doc1,
-                only_with_perms_in=["view_document"],
-                with_group_users=False,
-            ),
-        )
-
     @mock.patch("documents.models.Document.delete")
     def test_delete_documents_old_uuid_field(self, m) -> None:
         m.side_effect = Exception("Data too long for column 'transaction_id' at row 1")

src/documents/tests/test_classifier.py

@@ -3,7 +3,6 @@ import warnings
 from pathlib import Path
 from unittest import mock
 
-import numpy as np
 import pytest
 from django.conf import settings
 from django.test import TestCase
@@ -12,7 +11,6 @@ from django.test import override_settings
 from documents.classifier import ClassifierModelCorruptError
 from documents.classifier import DocumentClassifier
 from documents.classifier import IncompatibleClassifierVersionError
-from documents.classifier import _predict_with_threshold
 from documents.classifier import load_classifier
 from documents.models import Correspondent
 from documents.models import Document
@@ -627,103 +625,6 @@ class TestClassifier(DirectoriesMixin, TestCase):
         self.assertEqual(self.classifier.predict_storage_path(doc1.content), sp.pk)
         self.assertIsNone(self.classifier.predict_storage_path(doc2.content))
 
-    def test_predict_rejects_prediction_below_match_threshold(self) -> None:
-        """
-        GIVEN:
-            - Classifiers trained against test data with confident predictions
-        WHEN:
-            - CLASSIFIER_MATCH_THRESHOLD exceeds the model's confidence
-        THEN:
-            - Every predict_* method discards the match in favor of no match
-        """
-        c1 = Correspondent.objects.create(
-            name="c1",
-            matching_algorithm=Correspondent.MATCH_AUTO,
-        )
-        dt1 = DocumentType.objects.create(
-            name="dt1",
-            matching_algorithm=DocumentType.MATCH_AUTO,
-        )
-        sp1 = StoragePath.objects.create(
-            name="sp1",
-            matching_algorithm=StoragePath.MATCH_AUTO,
-        )
-
-        doc1 = Document.objects.create(
-            title="doc1",
-            content="this is a document from c1",
-            correspondent=c1,
-            document_type=dt1,
-            storage_path=sp1,
-            checksum="A",
-        )
-        Document.objects.create(
-            title="doc2",
-            content="this is a document from no one",
-            checksum="B",
-        )
-
-        self.classifier.train()
-
-        predictors = {
-            "correspondent": self.classifier.predict_correspondent,
-            "document_type": self.classifier.predict_document_type,
-            "storage_path": self.classifier.predict_storage_path,
-        }
-        # No real prediction can reach a confidence this high, so this
-        # isolates the threshold check from the model's actual output.
-        with override_settings(CLASSIFIER_MATCH_THRESHOLD=0.999999):
-            for name, predict in predictors.items():
-                with self.subTest(field=name):
-                    self.assertIsNone(predict(doc1.content))
-
-    def test_train_uses_balanced_sample_weight(self) -> None:
-        """
-        GIVEN:
-            - A training set with correspondents, document types and storage paths
-        WHEN:
-            - The classifier is trained
-        THEN:
-            - Each MLP classifier is fit with balanced sample weights, so that
-              over-represented classes don't dominate predictions
-        """
-        c1 = Correspondent.objects.create(
-            name="c1",
-            matching_algorithm=Correspondent.MATCH_AUTO,
-        )
-        dt1 = DocumentType.objects.create(
-            name="dt1",
-            matching_algorithm=DocumentType.MATCH_AUTO,
-        )
-        sp1 = StoragePath.objects.create(
-            name="sp1",
-            matching_algorithm=StoragePath.MATCH_AUTO,
-        )
-
-        Document.objects.create(
-            title="doc1",
-            content="this is a document from c1",
-            correspondent=c1,
-            document_type=dt1,
-            storage_path=sp1,
-            checksum="A",
-        )
-        Document.objects.create(
-            title="doc2",
-            content="this is a document from no one",
-            checksum="B",
-        )
-
-        with mock.patch(
-            "sklearn.utils.class_weight.compute_sample_weight",
-            return_value=None,
-        ) as mocked_compute_sample_weight:
-            self.classifier.train()
-
-        self.assertEqual(mocked_compute_sample_weight.call_count, 3)
-        for call in mocked_compute_sample_weight.call_args_list:
-            self.assertEqual(call.args[0], "balanced")
-
     def test_one_tag_predict(self) -> None:
         t1 = Tag.objects.create(name="t1", matching_algorithm=Tag.MATCH_AUTO, pk=12)
 
@@ -909,52 +810,6 @@ class TestClassifier(DirectoriesMixin, TestCase):
             load_classifier(raise_exception=True)
 
 
-class _StubProbaClassifier:
-    """
-    A fake scikit-learn classifier exposing just enough of the API for
-    `_predict_with_threshold`: `classes_` and `predict_proba`.
-    """
-
-    def __init__(self, classes: list[int], probabilities: list[float]) -> None:
-        self.classes_ = np.array(classes)
-        self._probabilities = np.array([probabilities])
-
-    def predict_proba(self, X) -> np.ndarray:
-        return self._probabilities
-
-
-@pytest.mark.parametrize(
-    ("classes", "probabilities", "threshold", "expected"),
-    [
-        # confident prediction above the threshold is returned
-        ([-1, 3], [0.1, 0.9], 0.6, 3),
-        # prediction below the threshold is discarded
-        ([-1, 3], [0.45, 0.55], 0.6, None),
-        # boundary: exactly at the threshold is accepted, not discarded
-        ([-1, 3], [0.4, 0.6], 0.6, 3),
-        # the winning class is the "no match" pseudo-class, regardless of its
-        # own confidence
-        ([-1, 3], [0.99, 0.01], 0.0, None),
-        # threshold of 0.0 disables the confidence check entirely
-        ([-1, 3], [0.45, 0.55], 0.0, 3),
-    ],
-)
-def test_predict_with_threshold(classes, probabilities, threshold, expected) -> None:
-    classifier = _StubProbaClassifier(classes, probabilities)
-    result = _predict_with_threshold(classifier, X=None, threshold=threshold)
-    assert result == expected
-
-
-def test_classifier_match_threshold_default() -> None:
-    """
-    GIVEN:
-        - No PAPERLESS_CLASSIFIER_MATCH_THRESHOLD environment variable is set
-    THEN:
-        - The classifier match threshold defaults to 0.6
-    """
-    assert settings.CLASSIFIER_MATCH_THRESHOLD == 0.6
-
-
 def test_preprocess_content() -> None:
     """
     GIVEN:

src/documents/tests/test_document_list_effective_content_annotation.py

@@ -1,457 +0,0 @@
-from __future__ import annotations
-
-from types import SimpleNamespace
-from typing import TYPE_CHECKING
-
-import pytest
-from django.db import connection
-from django.test.utils import CaptureQueriesContext
-from rest_framework import status
-
-from documents.models import Document
-from documents.tests.factories import DocumentFactory
-from documents.versioning import LATEST_VERSION_CONTENT_PREFETCH_ATTR
-from documents.versioning import has_prefetched_effective_content
-from documents.versioning import latest_version_content_prefetch
-from documents.views import DocumentViewSet
-
-if TYPE_CHECKING:
-    from rest_framework.test import APIClient
-
-
-class TestNeedsEffectiveContentAnnotation:
-    """
-    DocumentViewSet._needs_effective_content_annotation() decides whether
-    the effective_content correlated subquery is worth attaching to the
-    queryset at all -- see TestDocumentListEffectiveContentAnnotation below
-    for why. This only checks that decision's own logic (a plain query-param
-    membership test), not that Django/DRF's filtering machinery works.
-    """
-
-    @pytest.mark.parametrize(
-        ("params", "expected"),
-        [
-            ({}, False),
-            ({"ordering": "-added"}, False),
-            ({"tags__id__in": "1,2"}, False),
-            ({"search": ""}, False),
-            ({"search": "   "}, False),
-            ({"content__icontains": ""}, False),
-            ({"search": "foo"}, True),
-            ({"title_content": "foo"}, True),
-            ({"content__istartswith": "foo"}, True),
-            ({"content__iendswith": "foo"}, True),
-            ({"content__icontains": "foo"}, True),
-            ({"content__iexact": "foo"}, True),
-        ],
-    )
-    def test_detects_content_filter_params(
-        self,
-        params: dict[str, str],
-        expected: bool,  # noqa: FBT001
-    ) -> None:
-        """
-        GIVEN:
-            - A view bound to a request carrying the given query params
-        WHEN:
-            - Checking whether the effective_content annotation is needed
-        THEN:
-            - It is needed only for requests that actually filter on it
-        """
-        view = DocumentViewSet()
-        view.request = SimpleNamespace(query_params=params)
-
-        assert view._needs_effective_content_annotation() is expected
-
-
-class TestNeedsEffectiveContentPrefetch:
-    """
-    DocumentViewSet._needs_effective_content_prefetch() decides whether the
-    single-version content prefetch is worth attaching. It has to read the
-    `fields` param exactly the way get_serializer() does, or a request whose
-    response includes content ends up without the prefetch and pays
-    get_effective_content()'s per-instance fallback instead.
-    """
-
-    @pytest.mark.parametrize(
-        ("params", "expected"),
-        [
-            pytest.param({}, True, id="no-fields-param-keeps-every-field"),
-            pytest.param({"fields": ""}, True, id="blank-fields-keeps-every-field"),
-            pytest.param(
-                {"fields": "id,content"},
-                True,
-                id="content-among-requested-fields",
-            ),
-            pytest.param({"fields": "content"}, True, id="content-only"),
-            pytest.param({"fields": "id"}, False, id="content-not-requested"),
-            pytest.param(
-                {"fields": "id,title"},
-                False,
-                id="several-fields-without-content",
-            ),
-        ],
-    )
-    def test_detects_whether_content_can_reach_the_response(
-        self,
-        params: dict[str, str],
-        expected: bool,  # noqa: FBT001
-    ) -> None:
-        """
-        GIVEN:
-            - A view bound to a request carrying the given query params
-        WHEN:
-            - Checking whether the content prefetch is needed
-        THEN:
-            - It is needed exactly when get_serializer() would emit content,
-              which treats a blank `fields` the same as an absent one
-        """
-        view = DocumentViewSet()
-        view.request = SimpleNamespace(query_params=params)
-
-        assert view._needs_effective_content_prefetch() is expected
-
-
-@pytest.mark.django_db
-class TestDocumentListEffectiveContentAnnotation:
-    """
-    DocumentViewSet.get_queryset() only attaches the effective_content
-    correlated subquery when a request actually filters on it. Attaching it
-    unconditionally re-executes it once per candidate row before the page's
-    LIMIT is applied -- fine on SQLite/Postgres, but pathological on
-    MariaDB's default cardinality estimation for the root_document_id
-    self-join once candidate counts get large (see the root_document_id /
-    effective_content perf investigation).
-    """
-
-    def test_list_without_content_filter_skips_annotation_but_returns_latest_content(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document whose latest version has different content
-        WHEN:
-            - Listing documents with no search/content-filter param
-        THEN:
-            - The response still reflects the latest version's content
-            - The database never evaluates effective_content per row
-        """
-        root = DocumentFactory(content="old-root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="new-version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/documents/?fields=id,content")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert response.data["results"] == [
-            {"id": root.id, "content": "new-version-content"},
-        ]
-        assert not any(
-            "effective_content" in query["sql"] for query in ctx.captured_queries
-        )
-
-    @pytest.mark.parametrize(
-        "fields_param",
-        [
-            pytest.param("", id="blank-fields"),
-            pytest.param("id,content", id="content-requested"),
-        ],
-    )
-    def test_content_resolves_without_a_query_per_document(
-        self,
-        admin_client: APIClient,
-        fields_param: str,
-    ) -> None:
-        """
-        GIVEN:
-            - One versioned root document, then two more
-        WHEN:
-            - Listing documents with a `fields` param that keeps content
-        THEN:
-            - Every root's content resolves to its latest version's
-            - The query count does not grow with the number of documents,
-              i.e. a blank `fields` does not skip the prefetch and fall back
-              to loading each root's deferred version content
-        """
-        first = DocumentFactory(content="first-root-content")
-        DocumentFactory(
-            root_document=first,
-            version_index=1,
-            content="first-version-content",
-        )
-
-        with CaptureQueriesContext(connection) as one_document:
-            response = admin_client.get(f"/api/documents/?fields={fields_param}")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert [r["content"] for r in response.data["results"]] == [
-            "first-version-content",
-        ]
-
-        for index in range(2):
-            root = DocumentFactory(content=f"root-content-{index}")
-            DocumentFactory(
-                root_document=root,
-                version_index=1,
-                content=f"version-content-{index}",
-            )
-        with CaptureQueriesContext(connection) as three_documents:
-            response = admin_client.get(f"/api/documents/?fields={fields_param}")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert sorted(r["content"] for r in response.data["results"]) == [
-            "first-version-content",
-            "version-content-0",
-            "version-content-1",
-        ]
-        assert len(_get_document_queries(three_documents)) == len(
-            _get_document_queries(one_document),
-        )
-
-    def test_list_without_content_field_skips_prefetch_and_omits_content(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A versioned root document
-        WHEN:
-            - Listing documents without asking for content
-        THEN:
-            - Content is neither serialized nor resolved
-            - Nothing pays for the prefetch or the per-instance fallback
-        """
-        root = DocumentFactory(content="root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/documents/?fields=id")
-
-        assert response.status_code == status.HTTP_200_OK
-        assert response.data["results"] == [{"id": root.id}]
-        assert _get_effective_content_fallback_queries(ctx) == []
-        # Only the list query itself reads a content column: no extra query
-        # for the skipped prefetch, none for a per-instance fallback
-        content_queries = [
-            query
-            for query in ctx.captured_queries
-            if '"documents_document"."content"' in query["sql"]
-        ]
-        assert len(content_queries) == 1
-
-    def test_latest_version_content_prefetch_carries_only_the_newest_version(
-        self,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document with two versions
-        WHEN:
-            - Fetching the root through latest_version_content_prefetch()
-        THEN:
-            - The prefetch carries only the single newest version, not every
-              historical version's content (the whole point of not reusing
-              the metadata-only "versions" prefetch for this)
-        """
-        root = DocumentFactory(content="root-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="older-version-content",
-        )
-        DocumentFactory(
-            root_document=root,
-            version_index=2,
-            content="newest-version-content",
-        )
-
-        fetched_root = (
-            Document.objects.filter(pk=root.pk)
-            .prefetch_related(
-                latest_version_content_prefetch(),
-            )
-            .get()
-        )
-
-        latest = getattr(fetched_root, LATEST_VERSION_CONTENT_PREFETCH_ATTR)
-        assert [v.content for v in latest] == ["newest-version-content"]
-
-
-class TestHasPrefetchedEffectiveContent:
-    """
-    DocumentSerializer.to_representation() only calls get_effective_content()
-    when has_prefetched_effective_content() says it's cheap -- otherwise a
-    caller that never set up an annotation or prefetch (TrashView,
-    GlobalSearchView, which build their own querysets and don't display
-    content at all) would pay for a per-instance query nobody asked for.
-    """
-
-    def test_false_with_no_annotation_or_prefetch(self) -> None:
-        """
-        GIVEN:
-            - A document the ORM never annotated or prefetched for
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is not, so the serializer must leave it alone
-        """
-        document = DocumentFactory.build()
-
-        assert has_prefetched_effective_content(document) is False
-
-    def test_true_with_effective_content_annotation(self) -> None:
-        """
-        GIVEN:
-            - A document carrying the queryset's effective_content annotation
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is, straight off the annotation
-        """
-        document = DocumentFactory.build()
-        document.effective_content = "resolved"
-
-        assert has_prefetched_effective_content(document) is True
-
-    def test_true_with_lean_prefetch_attr_even_when_empty(self) -> None:
-        """
-        GIVEN:
-            - A document the lean content prefetch ran for, finding no versions
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is: an empty prefetch is an answer, not a missing one
-        """
-        document = DocumentFactory.build()
-        setattr(document, LATEST_VERSION_CONTENT_PREFETCH_ATTR, [])
-
-        assert has_prefetched_effective_content(document) is True
-
-    def test_true_with_metadata_versions_prefetch_cache(self) -> None:
-        """
-        GIVEN:
-            - A document carrying only the metadata "versions" prefetch
-        WHEN:
-            - Asking whether its effective content is already resolved
-        THEN:
-            - It is, via get_effective_content()'s prefetch-cache branch
-        """
-        document = DocumentFactory.build()
-        document._prefetched_objects_cache = {"versions": []}
-
-        assert has_prefetched_effective_content(document) is True
-
-
-def _get_document_queries(
-    ctx: CaptureQueriesContext,
-) -> list[dict[str, str]]:
-    """
-    The queries a list request spends on the documents themselves, i.e.
-    everything but the one-time django_content_type lookup guardian's
-    permission filtering makes. That lookup is process-cached, and the
-    autouse fixture in conftest clears the cache before every test, so it
-    lands in whichever request happens to run first and never repeats --
-    counting it makes a request look like it costs one query more than the
-    identical request after it.
-    """
-    return [q for q in ctx.captured_queries if '"django_content_type"' not in q["sql"]]
-
-
-def _get_effective_content_fallback_queries(
-    ctx: CaptureQueriesContext,
-) -> list[dict[str, str]]:
-    """
-    Document.get_effective_content()'s per-instance fallback (no annotation,
-    no prefetch) is a `.values_list("content", flat=True).first()` query --
-    a SELECT of just the content column. Distinct from get_versions()'s own,
-    unrelated per-instance metadata query (id/checksum/added/etc, no
-    content) run to build the "versions" response field, which isn't part
-    of what this test file covers.
-    """
-    return [
-        q
-        for q in ctx.captured_queries
-        if q["sql"].startswith('SELECT "documents_document"."content" FROM')
-    ]
-
-
-@pytest.mark.django_db
-class TestTrashAndGlobalSearchEffectiveContentIsNeverPerInstance:
-    """
-    TrashView and GlobalSearchView serialize Document instances with
-    DocumentSerializer too, but build their querysets independently of
-    DocumentViewSet.get_queryset(). TrashView doesn't display content at all,
-    so it keeps the document's own unresolved content; GlobalSearchView
-    annotates effective_content itself, so it shows the latest version's.
-    Neither should ever fall back to a per-instance query.
-    """
-
-    def test_trash_list_shows_unresolved_content_with_no_extra_query(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A trashed root document whose own content differs from what a
-              version would have had (also trashed, deletion cascades)
-        WHEN:
-            - Listing trash
-        THEN:
-            - The response shows the document's own content
-            - Nothing ever queries for versions to resolve it
-        """
-        root = DocumentFactory(content="own-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-        root.delete()
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get("/api/trash/")
-
-        assert response.status_code == status.HTTP_200_OK
-        [result] = [r for r in response.data["results"] if r["id"] == root.id]
-        assert result["content"] == "own-content"
-        assert _get_effective_content_fallback_queries(ctx) == []
-
-    def test_global_search_db_only_shows_latest_version_content_with_no_extra_query(
-        self,
-        admin_client: APIClient,
-    ) -> None:
-        """
-        GIVEN:
-            - A root document, findable by title, whose own content differs
-              from its latest version's
-        WHEN:
-            - Using the global search endpoint's db_only mode
-        THEN:
-            - The response shows the latest version's content, resolved by
-              GlobalSearchView's own effective_content annotation
-            - There is no per-instance fallback query
-        """
-        root = DocumentFactory(title="findme", content="own-content")
-        DocumentFactory(
-            root_document=root,
-            version_index=1,
-            content="version-content",
-        )
-
-        with CaptureQueriesContext(connection) as ctx:
-            response = admin_client.get(
-                "/api/search/?query=findme&db_only=true",
-            )
-
-        assert response.status_code == status.HTTP_200_OK
-        [result] = [d for d in response.data["documents"] if d["id"] == root.id]
-        assert result["content"] == "version-content"
-        assert _get_effective_content_fallback_queries(ctx) == []

src/documents/tests/test_document_model.py

@@ -110,7 +110,7 @@ class TestDocument(TestCase):
             checksum="checksum",
             mime_type="application/pdf",
         )
-        version = Document.objects.create(
+        Document.objects.create(
             root_document=root,
             correspondent=root.correspondent,
             title="Version",
@@ -124,10 +124,6 @@ class TestDocument(TestCase):
         self.assertEqual(Document.objects.count(), 0)
         self.assertEqual(Document.deleted_objects.count(), 2)
 
-        root.restore(strict=False)
-
-        self.assertTrue(Document.objects.filter(pk=version.pk).exists())
-
     def test_file_name(self) -> None:
         doc = Document(
             mime_type="application/pdf",

src/documents/tests/test_get_public_filename_nfc.py

@@ -0,0 +1,48 @@
+import unicodedata
+from datetime import date
+
+import pytest
+
+from documents.models import Correspondent
+from documents.models import Document
+
+
+@pytest.mark.django_db
+class TestGetPublicFilenameNfc:
+    def test_normalizes_nfd_title_to_nfc(self) -> None:
+        nfd_title = unicodedata.normalize("NFD", "Gehaltserhöhung")
+        assert not unicodedata.is_normalized("NFC", nfd_title)
+
+        doc = Document(
+            mime_type="application/pdf",
+            title=nfd_title,
+            created=date(2025, 10, 17),
+        )
+
+        result = doc.get_public_filename()
+
+        assert unicodedata.is_normalized("NFC", result)
+        assert (
+            result
+            == "2025-10-17 "
+            + unicodedata.normalize(
+                "NFC",
+                nfd_title,
+            )
+            + ".pdf"
+        )
+
+    def test_normalizes_nfd_correspondent_name_to_nfc(self) -> None:
+        nfd_name = unicodedata.normalize("NFD", "Müller GmbH")
+        correspondent = Correspondent.objects.create(name=nfd_name)
+
+        doc = Document.objects.create(
+            mime_type="application/pdf",
+            title="Rechnung",
+            created=date(2025, 10, 17),
+            correspondent=correspondent,
+        )
+
+        result = doc.get_public_filename()
+
+        assert unicodedata.is_normalized("NFC", result)

src/documents/tests/test_management_consumer.py

@@ -136,23 +136,6 @@ def wait_for_mock_call(
     return False
 
 
-def sleep_past_stability(
-    owner: FileStabilityTracker | ConsumerThread,
-    *,
-    windows: float = 1.5,
-) -> None:
-    """
-    Block until a tracked file's stability window has certainly elapsed.
-
-    Args:
-        owner: The tracker, or the consumer thread running one, whose
-               configured stability delay sets the wait.
-        windows: How many stability windows to wait, giving slop for a slow
-                 or loaded test runner.
-    """
-    sleep(owner.stability_delay * windows)
-
-
 class TestTrackedFile:
     """Tests for the TrackedFile dataclass."""
 
@@ -278,56 +261,6 @@ class TestFileStabilityTracker:
         assert len(stable) == 0
         assert stability_tracker.pending_count == 1
 
-    def test_get_stable_files_skips_empty_file(
-        self,
-        stability_tracker: FileStabilityTracker,
-        tmp_path: Path,
-    ) -> None:
-        """
-        GIVEN:
-            - A zero byte file, tracked and past its stability delay
-        WHEN:
-            - Stable files are collected
-        THEN:
-            - The file is not yielded for consumption
-            - The file is dropped from tracking rather than held, so an
-              abandoned placeholder does not keep the watch loop awake
-        """
-        empty = tmp_path / "scan.pdf"
-        empty.write_bytes(b"")
-        stability_tracker.track(empty, Change.added)
-        sleep_past_stability(stability_tracker)
-
-        stable = list(stability_tracker.get_stable_files())
-
-        assert stable == []
-        assert stability_tracker.pending_count == 0
-
-    def test_empty_file_is_yielded_once_content_arrives(
-        self,
-        stability_tracker: FileStabilityTracker,
-        tmp_path: Path,
-    ) -> None:
-        """
-        GIVEN:
-            - A zero byte file which was dropped from tracking while empty
-        WHEN:
-            - The writer fills the file and a new event re-tracks it
-        THEN:
-            - The file is yielded for consumption once it is stable
-        """
-        target = tmp_path / "scan.pdf"
-        target.write_bytes(b"")
-        stability_tracker.track(target, Change.added)
-        sleep_past_stability(stability_tracker)
-        assert list(stability_tracker.get_stable_files()) == []
-
-        target.write_bytes(b"%PDF-1.4 content")
-        stability_tracker.track(target, Change.modified)
-        sleep_past_stability(stability_tracker)
-
-        assert list(stability_tracker.get_stable_files()) == [target]
-
     def test_get_stable_files_deleted_during_check(self, temp_file: Path) -> None:
         """Test deleted file is not returned during stability check."""
         tracker = FileStabilityTracker(stability_delay=0.1)
@@ -946,51 +879,6 @@ class TestCommandWatch:
 
         mock_consume_file_delay.apply_async.assert_called()
 
-    def test_scanner_placeholder_is_not_consumed_while_empty(
-        self,
-        consumption_dir: Path,
-        sample_pdf: Path,
-        mock_consume_file_delay: MagicMock,
-        start_consumer: Callable[..., ConsumerThread],
-    ) -> None:
-        """
-        GIVEN:
-            - A scanner which creates a zero byte placeholder and only writes
-              the page some time later (GH discussion #13969)
-        WHEN:
-            - The placeholder sits untouched well past the stability delay
-            - The scanner then writes the real content
-        THEN:
-            - The empty placeholder is never queued, as it could only fail
-              with "Unsupported mime type inode/x-empty"
-            - The file is queued exactly once, when the content lands
-        """
-        thread = start_consumer(stability_delay=0.2)
-
-        target = consumption_dir / "scan.pdf"
-        target.write_bytes(b"")  # the scanner's placeholder
-
-        # Well past the stability delay: the old behaviour queued it here.
-        sleep_past_stability(thread, windows=5)
-        if thread.exception:
-            raise thread.exception
-        assert mock_consume_file_delay.apply_async.call_count == 0
-
-        shutil.copy(sample_pdf, target)  # the scanner finishes the page
-
-        assert wait_for_mock_call(
-            mock_consume_file_delay.apply_async,
-            timeout_s=5.0,
-        )
-        if thread.exception:
-            raise thread.exception
-
-        assert mock_consume_file_delay.apply_async.call_count == 1
-        queued_doc = mock_consume_file_delay.apply_async.call_args.kwargs["kwargs"][
-            "input_doc"
-        ]
-        assert queued_doc.original_file.name == "scan.pdf"
-
     def test_ignores_macos_files(
         self,
         consumption_dir: Path,

src/documents/tests/test_matching_nfc.py

@@ -0,0 +1,80 @@
+import unicodedata
+
+import pytest
+
+from documents.data_models import ConsumableDocument
+from documents.data_models import DocumentSource
+from documents.matching import consumable_document_matches_workflow
+from documents.matching import existing_document_matches_workflow
+from documents.models import Document
+from documents.models import Workflow
+from documents.models import WorkflowTrigger
+
+
+@pytest.mark.django_db
+class TestMatchingNfcNormalization:
+    def test_consumable_document_filename_nfd_matches_nfc_pattern(
+        self,
+        tmp_path,
+    ) -> None:
+        """
+        GIVEN:
+            - A file on disk whose name is NFD-normalized
+            - A workflow trigger filename filter typed as NFC
+        WHEN:
+            - The consumable document is checked against the trigger
+        THEN:
+            - It matches, because both sides are normalized before comparing
+        """
+        nfd_name = unicodedata.normalize("NFD", "Gehaltserhöhung.pdf")
+        nfc_pattern = unicodedata.normalize("NFC", "*Gehaltserhöhung*")
+        assert nfd_name != unicodedata.normalize("NFC", nfd_name)
+
+        file_path = tmp_path / nfd_name
+        file_path.write_bytes(b"%PDF-1.4 test")
+
+        document = ConsumableDocument(
+            source=DocumentSource.ConsumeFolder,
+            original_file=file_path,
+        )
+        trigger = WorkflowTrigger(
+            type=WorkflowTrigger.WorkflowTriggerType.CONSUMPTION,
+            filter_filename=nfc_pattern,
+            sources=[],
+        )
+
+        matched, reason = consumable_document_matches_workflow(document, trigger)
+
+        assert matched, reason
+
+    def test_existing_document_filename_nfd_matches_nfc_pattern(self) -> None:
+        """
+        GIVEN:
+            - A Document whose original_filename is NFD-normalized (e.g. from
+              before normalization was applied at consumption time)
+            - A workflow trigger filename filter typed as NFC
+        WHEN:
+            - The document is checked against the trigger
+        THEN:
+            - It matches, because both sides are normalized before comparing
+        """
+        nfd_name = unicodedata.normalize("NFD", "Gehaltserhöhung.pdf")
+        nfc_pattern = unicodedata.normalize("NFC", "*Gehaltserhöhung*")
+
+        document = Document.objects.create(
+            title="Test",
+            content="content",
+            checksum="checksum",
+            mime_type="application/pdf",
+            original_filename=nfd_name,
+        )
+        workflow = Workflow.objects.create(name="Test workflow", order=0)
+        trigger = WorkflowTrigger.objects.create(
+            type=WorkflowTrigger.WorkflowTriggerType.DOCUMENT_ADDED,
+            filter_filename=nfc_pattern,
+        )
+        workflow.triggers.add(trigger)
+
+        matched, reason = existing_document_matches_workflow(document, trigger)
+
+        assert matched, reason

src/documents/tests/test_normalize_unicode.py

@@ -0,0 +1,6 @@
+from documents.utils import normalize_unicode
+
+
+class TestNormalizeUnicode:
+    def test_none_passes_through(self) -> None:
+        assert normalize_unicode(None) is None

src/documents/tests/test_tag_hierarchy.py

@@ -2,7 +2,6 @@ from unittest import mock
 
 from django.contrib.auth.models import Permission
 from django.contrib.auth.models import User
-from rest_framework import status
 from rest_framework.test import APITestCase
 
 from documents import bulk_edit
@@ -109,44 +108,6 @@ class TestTagHierarchy(DirectoriesMixin, APITestCase):
         self.document.refresh_from_db()
         assert self.document.tags.count() == 0
 
-    def test_remove_inbox_tags_removes_nested_children(self) -> None:
-        inbox = Tag.objects.create(name="Inbox", is_inbox_tag=True)
-        nested = Tag.objects.create(name="Nested", tn_parent=inbox)
-        self.document.add_nested_tags([nested])
-
-        resp = self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"title": "new title", "remove_inbox_tags": True},
-            format="json",
-        )
-        assert resp.status_code == status.HTTP_200_OK
-        self.document.refresh_from_db()
-        assert self.document.tags.count() == 0
-
-        # A subsequent save must not re-add the inbox tag as an ancestor
-        resp = self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"title": "another title", "tags": [], "remove_inbox_tags": True},
-            format="json",
-        )
-        assert resp.status_code == status.HTTP_200_OK
-        self.document.refresh_from_db()
-        assert self.document.tags.count() == 0
-
-    def test_remove_inbox_tags_keeps_inbox_when_nested_child_added(self) -> None:
-        inbox = Tag.objects.create(name="Inbox", is_inbox_tag=True)
-        nested = Tag.objects.create(name="Nested", tn_parent=inbox)
-        self.document.add_nested_tags([inbox])
-
-        self.client.patch(
-            f"/api/documents/{self.document.pk}/",
-            {"tags": [nested.pk], "remove_inbox_tags": True},
-            format="json",
-        )
-        self.document.refresh_from_db()
-        tags = set(self.document.tags.values_list("pk", flat=True))
-        assert tags == {inbox.pk, nested.pk}
-
     def test_bulk_edit_respects_hierarchy(self) -> None:
         bulk_edit.add_tag([self.document.pk], self.child.pk)
         self.document.refresh_from_db()

src/documents/tests/test_views.py

@@ -32,7 +32,6 @@ from documents.signals.handlers import update_llm_suggestions_cache
 from documents.tests.utils import DirectoriesMixin
 from documents.tests.utils import read_streaming_response
 from paperless.models import ApplicationConfiguration
-from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 
 
@@ -738,38 +737,6 @@ class TestAISuggestions(DirectoriesMixin, TestCase):
             get_llm_suggestion_cache(self.document.pk, backend="openai-like"),
         )
 
-    @patch("documents.views.get_ai_document_classification")
-    @override_settings(
-        AI_ENABLED=True,
-        LLM_BACKEND="openai-like",
-    )
-    def test_ai_suggestions_with_llm_provider_error(
-        self,
-        mock_get_ai_classification,
-    ) -> None:
-        mock_get_ai_classification.side_effect = LLMProviderError(
-            "confidential provider response",
-        )
-
-        self.client.force_login(user=self.user)
-        response = self.client.get(
-            f"/api/documents/{self.document.pk}/ai_suggestions/",
-        )
-
-        self.assertEqual(response.status_code, status.HTTP_502_BAD_GATEWAY)
-        self.assertEqual(
-            response.json(),
-            {
-                "ai": [
-                    "AI backend rejected the request. Check logs for details.",
-                ],
-            },
-        )
-        self.assertNotIn("confidential provider response", response.content.decode())
-        self.assertIsNone(
-            get_llm_suggestion_cache(self.document.pk, backend="openai-like"),
-        )
-
     @patch("documents.views.get_ai_document_classification")
     @override_settings(
         AI_ENABLED=True,

src/documents/utils.py

@@ -1,6 +1,7 @@
 import hashlib
 import logging
 import shutil
+import unicodedata
 from collections.abc import Callable
 from collections.abc import Iterable
 from collections.abc import Iterator
@@ -31,6 +32,25 @@ def identity(iterable: Iterable[_T]) -> Iterable[_T]:
     return iterable
 
 
+def normalize_unicode(value: str | None) -> str | None:
+    """
+    Normalize a string to Unicode NFC form, or return None unchanged.
+
+    This is the single normalization pass for any user- or filesystem-supplied
+    text that ends up in a filename, path, or is compared/matched against one
+    (titles, correspondent/tag/type names, uploaded filenames, workflow and
+    mail rule filename/path filters). Composed (NFC) and decomposed (NFD)
+    forms of the same visible text are different byte sequences, which breaks
+    exact comparisons and filesystem lookups even though the text looks
+    identical. Always normalize through this function rather than calling
+    unicodedata.normalize() directly, so every call site agrees on the same
+    form.
+    """
+    if value is None:
+        return None
+    return unicodedata.normalize("NFC", value)
+
+
 class QuerySetStream(Generic[_M]):
     """Stream a QuerySet via .iterator(chunk_size=...) instead of
     materializing it (plus any prefetch caches) all at once, while still

src/documents/versioning.py

@@ -7,12 +7,9 @@ from typing import Any
 
 from django.db.models import F
 from django.db.models import OuterRef
-from django.db.models import Prefetch
 from django.db.models import QuerySet
 from django.db.models import Subquery
-from django.db.models import Window
 from django.db.models.functions import Coalesce
-from django.db.models.functions import RowNumber
 
 from documents.models import Document
 
@@ -49,68 +46,6 @@ def annotate_effective_content(documents: QuerySet[Document]) -> QuerySet[Docume
     )
 
 
-LATEST_VERSION_CONTENT_PREFETCH_ATTR = "_latest_version_content_prefetch"
-
-
-def latest_version_content_prefetch() -> Prefetch:
-    """
-    A Prefetch for Document.versions scoped to just the newest version's
-    content, for get_effective_content()'s fallback when no SQL annotation
-    is present.
-
-    Deliberately not merged into a metadata-only "versions" prefetch (the one
-    used for the serialized versions list): that one fetches every historical
-    version of every document, and pulling full OCR content for versions
-    nobody will read wastes DB transfer/memory at scale. This one is windowed
-    down to a single row per root, then bounded by Prefetch's own IN-list to
-    whatever page/result set it's attached to -- one cheap bulk query total,
-    not one per document and not one per version.
-    """
-    return Prefetch(
-        "versions",
-        queryset=(
-            Document.objects.filter(
-                root_document_id__isnull=False,
-                deleted_at__isnull=True,
-            )
-            .annotate(
-                rn=Window(
-                    RowNumber(),
-                    partition_by=F("root_document_id"),
-                    order_by=[
-                        F("version_index").desc(nulls_last=True),
-                        F("id").desc(),
-                    ],
-                ),
-            )
-            .filter(rn=1)
-            .only("id", "root_document_id", "content")
-        ),
-        to_attr=LATEST_VERSION_CONTENT_PREFETCH_ATTR,
-    )
-
-
-def has_prefetched_effective_content(document: Document) -> bool:
-    """
-    True if document.get_effective_content() can answer without an extra
-    per-instance query -- an SQL ``effective_content`` annotation, the lean
-    latest_version_content_prefetch(), or the metadata-only "versions"
-    prefetch is already present on the instance.
-
-    Callers that haven't set any of those up (e.g. views that build their
-    own querysets independently of DocumentViewSet.get_queryset(), like
-    TrashView or GlobalSearchView) intentionally don't pay for version-aware
-    content resolution -- see DocumentSerializer.to_representation(), which
-    uses this to decide whether to call get_effective_content() at all.
-    """
-    if hasattr(document, "effective_content"):
-        return True
-    if getattr(document, LATEST_VERSION_CONTENT_PREFETCH_ATTR, None) is not None:
-        return True
-    prefetched_cache = getattr(document, "_prefetched_objects_cache", None)
-    return isinstance(prefetched_cache, dict) and "versions" in prefetched_cache
-
-
 def sort_versions_newest_first(documents: list[Document]) -> list[Document]:
     """
     Same sorting as versions_newest_first()

src/documents/views.py

@@ -36,6 +36,7 @@ from django.db.migrations.recorder import MigrationRecorder
 from django.db.models import Avg
 from django.db.models import Case
 from django.db.models import Count
+from django.db.models import F
 from django.db.models import IntegerField
 from django.db.models import Max
 from django.db.models import Model
@@ -136,14 +137,12 @@ from documents.filters import CustomFieldFilterSet
 from documents.filters import DocumentFilterSet
 from documents.filters import DocumentsOrderingFilter
 from documents.filters import DocumentTypeFilterSet
-from documents.filters import EffectiveContentFilter
 from documents.filters import PaperlessTaskFilterSet
 from documents.filters import PermittedObjectsFilter
 from documents.filters import ShareLinkBundleFilterSet
 from documents.filters import ShareLinkFilterSet
 from documents.filters import StoragePathFilterSet
 from documents.filters import TagFilterSet
-from documents.filters import TitleContentFilter
 from documents.mail import EmailAttachment
 from documents.mail import send_email
 from documents.matching import match_correspondents
@@ -180,7 +179,7 @@ from documents.permissions import has_perms_owner_aware
 from documents.permissions import has_system_status_permission
 from documents.permissions import permitted_document_ids
 from documents.permissions import permitted_object_ids
-from documents.permissions import set_permissions_for_objects
+from documents.permissions import set_permissions_for_object
 from documents.permissions import user_is_unrestricted
 from documents.plugins.date_parsing import get_date_parser
 from documents.schema import generate_object_with_permissions_schema
@@ -232,12 +231,12 @@ from documents.tasks import sanity_check
 from documents.tasks import train_classifier
 from documents.tasks import update_document_parent_tags
 from documents.utils import get_boolean
+from documents.utils import normalize_unicode
 from documents.versioning import VersionResolutionError
 from documents.versioning import annotate_effective_content
 from documents.versioning import get_latest_version_for_root
 from documents.versioning import get_request_version_param
 from documents.versioning import get_root_document
-from documents.versioning import latest_version_content_prefetch
 from documents.versioning import resolve_requested_version_for_root
 from documents.versioning import versions_newest_first
 from paperless import version
@@ -254,7 +253,6 @@ from paperless.views import StandardPagination
 from paperless_ai.ai_classifier import get_ai_document_classification
 from paperless_ai.ai_classifier import get_llm_output_language
 from paperless_ai.chat import stream_chat_with_documents
-from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 from paperless_ai.matching import extract_unmatched_names
 from paperless_ai.matching import match_correspondents_by_name
@@ -1086,59 +1084,12 @@ class DocumentViewSet(
             ],
         }
 
-    @classmethod
-    def _content_filter_params(cls) -> tuple[str, ...]:
-        """
-        Query params whose filtering needs effective_content evaluated in SQL
-        against every candidate row -- see
-        _needs_effective_content_annotation(). Derived rather than
-        hand-maintained so a new content-filtering param counts automatically.
-        """
-        params = [
-            name
-            for name, f in DocumentFilterSet.declared_filters.items()
-            if isinstance(f, (TitleContentFilter, EffectiveContentFilter))
-        ]
-        if "effective_content" in cls.search_fields:
-            params.append(SearchFilter().search_param)
-        return tuple(params)
-
-    def _needs_effective_content_annotation(self) -> bool:
-        # effective_content is a per-row correlated subquery resolving each
-        # document's latest version. Filtering *on* it forces the database to
-        # evaluate it for every candidate row before reaching the LIMIT, which
-        # the root_document_id self-join makes pathological on MariaDB
-        # specifically once real candidate counts get large; otherwise the
-        # "versions" prefetch + Document.get_effective_content() resolves only
-        # the page that survives pagination. Every param here is deprecated in
-        # favor of the Tantivy-backed search endpoint (see filters.py's
-        # TitleContentFilter/EffectiveContentFilter docs), so pay that cost
-        # only when one is actually used. Blank values don't count, matching
-        # how those filters themselves no-op on them -- an empty `?search=`
-        # applies no predicate.
-        params = self.request.query_params
-        return any(
-            params.get(param, "").strip() for param in self._content_filter_params()
-        )
-
-    def _requested_fields(self) -> list[str] | None:
-        # The sparse-fieldset `fields` param, as DynamicFieldsModelSerializer
-        # wants it: None means "no restriction, serialize everything", which
-        # a blank value means too. get_queryset() and get_serializer() both
-        # branch on this, and they have to read it identically -- a queryset
-        # that skips the content prefetch for a response that still
-        # serializes content reintroduces get_effective_content()'s
-        # per-instance fallback.
-        fields_param = self.request.query_params.get("fields")
-        return fields_param.split(",") if fields_param else None
-
-    def _needs_effective_content_prefetch(self) -> bool:
-        # The prefetch spares get_effective_content() a per-instance fallback
-        # query, but only earns itself when content can reach the response.
-        fields = self._requested_fields()
-        return fields is None or "content" in fields
-
     def get_queryset(self):
+        latest_version_content = Subquery(
+            versions_newest_first(
+                Document.objects.filter(root_document=OuterRef("pk")),
+            ).values("content")[:1],
+        )
         # A correlated subquery avoids the LEFT JOIN + Count() this used to
         # be, which forced a GROUP BY aggregate over every matching document
         # before the query could even be sorted or limited.
@@ -1158,43 +1109,40 @@ class DocumentViewSet(
         # ObjectFilter.filter(). A blanket .distinct() here forces the
         # database to fully sort and dedupe every visible document before
         # it can apply LIMIT, which is disastrous at scale.
-        prefetches = [
+        return (
-            Prefetch(
-                "versions",
-                queryset=Document.objects.only(
-                    "id",
-                    "added",
-                    "checksum",
-                    "version_label",
-                    "root_document_id",
-                    "version_index",
-                ),
-            ),
-            "tags",
-            Prefetch(
-                "custom_fields",
-                queryset=CustomFieldInstance.objects.select_related("field"),
-            ),
-            # NotesSerializer nests the author, this avoids query per note
-            Prefetch("notes", queryset=Note.objects.select_related("user")),
-        ]
-        if self._needs_effective_content_prefetch():
-            prefetches.append(latest_version_content_prefetch())
-        queryset = (
             Document.objects.filter(root_document__isnull=True)
             .order_by("-created", "-id")
+            .annotate(effective_content=Coalesce(latest_version_content, F("content")))
             .annotate(num_notes=Coalesce(note_count, 0))
             .select_related("correspondent", "storage_path", "document_type", "owner")
-            .prefetch_related(*prefetches)
+            .prefetch_related(
+                Prefetch(
+                    "versions",
+                    queryset=Document.objects.only(
+                        "id",
+                        "added",
+                        "checksum",
+                        "version_label",
+                        "root_document_id",
+                        "version_index",
+                    ),
+                ),
+                "tags",
+                Prefetch(
+                    "custom_fields",
+                    queryset=CustomFieldInstance.objects.select_related("field"),
+                ),
+                # NotesSerializer nests the author, this avoids query per note
+                Prefetch("notes", queryset=Note.objects.select_related("user")),
+            )
         )
-        if self._needs_effective_content_annotation():
-            queryset = annotate_effective_content(queryset)
-        return queryset
 
     def get_serializer(self, *args, **kwargs):
+        fields_param = self.request.query_params.get("fields", None)
+        fields = fields_param.split(",") if fields_param else None
         truncate_content = self.request.query_params.get("truncate_content", "False")
         kwargs.setdefault("context", self.get_serializer_context())
-        kwargs.setdefault("fields", self._requested_fields())
+        kwargs.setdefault("fields", fields)
         kwargs.setdefault("truncate_content", truncate_content.lower() in ["true", "1"])
         try:
             full_perms = get_boolean(
@@ -1656,22 +1604,6 @@ class DocumentViewSet(
                     {"ai": [_("AI backend request timed out.")]},
                     status=status.HTTP_503_SERVICE_UNAVAILABLE,
                 )
-            except LLMProviderError:
-                logger.exception(
-                    "AI backend rejected the request for document %s",
-                    doc.pk,
-                )
-                return Response(
-                    {
-                        "ai": [
-                            _(
-                                "AI backend rejected the request. "
-                                "Check logs for details.",
-                            ),
-                        ],
-                    },
-                    status=status.HTTP_502_BAD_GATEWAY,
-                )
             set_llm_suggestions_cache(
                 doc.pk,
                 llm_suggestions,
@@ -2137,6 +2069,7 @@ class DocumentViewSet(
 
         try:
             doc_name, doc_data = serializer.validated_data.get("document")
+            doc_name = normalize_unicode(doc_name)
             version_label = serializer.validated_data.get("version_label")
 
             t = int(mktime(datetime.now().timetuple()))
@@ -3403,7 +3336,7 @@ class PostDocumentView(GenericAPIView[Any]):
         serializer.is_valid(raise_exception=True)
 
         doc_name, doc_data = serializer.validated_data.get("document")
-        doc_name = normalize("NFC", doc_name)
+        doc_name = normalize_unicode(doc_name)
         correspondent_id = serializer.validated_data.get("correspondent")
         document_type_id = serializer.validated_data.get("document_type")
         storage_path_id = serializer.validated_data.get("storage_path")
@@ -5019,12 +4952,12 @@ class BulkEditObjectsView(PassUserMixin):
                     qs_owner_update.update(owner=owner)
 
                 if "permissions" in serializer.validated_data:
-                    set_permissions_for_objects(
+                    for obj in qs:
-                        permissions=permissions,
+                        set_permissions_for_object(
-                        model=object_class,
+                            permissions=permissions,
-                        pks=qs.values_list("pk", flat=True),
+                            object=obj,
-                        merge=merge,
+                            merge=merge,
-                    )
+                        )
 
             except Exception as e:
                 logger.warning(
@@ -5506,10 +5439,7 @@ class TrashView(ListModelMixin, PassUserMixin):
 
     model = Document
 
-    # A version is listed separately only when its root is not in the trash.
+    queryset = Document.deleted_objects.all()
-    queryset = Document.deleted_objects.exclude(
-        root_document_id__in=Document.deleted_objects.values("id"),
-    )
 
     def get(self, request: Request, format: str | None = None) -> Response:
         self.serializer_class = DocumentSerializer
@@ -5540,15 +5470,7 @@ class TrashView(ListModelMixin, PassUserMixin):
             return HttpResponseForbidden("Insufficient permissions")
         action = serializer.validated_data.get("action")
         if action == "restore":
-            restored = list(self.get_queryset().filter(id__in=doc_ids))
+            restored = list(Document.deleted_objects.filter(id__in=doc_ids))
-            if len(restored) != len(doc_ids):
-                raise ValidationError(
-                    {
-                        "documents": [
-                            "Restore the root document instead of one of its versions.",
-                        ],
-                    },
-                )
             for doc in restored:
                 doc.restore(strict=False)
             if restored:

src/paperless/settings/__init__.py

@@ -96,13 +96,6 @@ MODEL_FILE = get_path_from_env(
     "PAPERLESS_MODEL_FILE",
     DATA_DIR / "classification_model.pickle",
 )
-
-# Minimum confidence (0.0-1.0) for the ML classifier to assign a correspondent,
-# document type, or storage path. 0.0 disables the threshold.
-CLASSIFIER_MATCH_THRESHOLD: Final[float] = get_float_from_env(
-    "PAPERLESS_CLASSIFIER_MATCH_THRESHOLD",
-    0.6,
-)
 LLM_INDEX_DIR = DATA_DIR / "llm_index"
 LLM_INDEX_LOCK = LLM_INDEX_DIR / "index.lock"
 # Cross-process read/write lock guarding the LLM index compaction/migration

src/paperless_ai/ai_classifier.py

@@ -4,24 +4,21 @@ from django.conf import settings
 from django.contrib.auth.models import User
 
 from documents.models import Document
-from documents.permissions import permitted_object_ids
+from documents.permissions import get_objects_for_user_owner_aware
-from documents.permissions import restrict_queryset_to_visible
-from documents.permissions import user_is_unrestricted
 from paperless.config import AIConfig
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import TaxonomyChoiceDict
 from paperless_ai.base_model import classification_suggestions_to_model
 from paperless_ai.client import AIClient
 from paperless_ai.db import db_connection_released
+from paperless_ai.indexing import _node_document_ids
 from paperless_ai.indexing import retrieve_similar_nodes
 from paperless_ai.indexing import truncate_content
 from paperless_ai.prompts.context import ClassificationPromptContext
 from paperless_ai.prompts.context import LocalizationPromptContext
 from paperless_ai.prompts.context import RagContextPromptContext
 from paperless_ai.prompts.render import render_prompt
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
-from paperless_ai.taxonomy import _node_document_weights
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import empty_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
@@ -40,48 +37,6 @@ logger = logging.getLogger("paperless_ai.rag_classifier")
 TAXONOMY_CANDIDATE_TOP_K = 15
 
 
-def _fulltext_similar_documents(
-    document: Document,
-    user: User | None,
-    top_k: int,
-) -> list[SimilarDocument]:
-    """Rank-based fallback when no embedding backend is configured. Uses
-    Tantivy's "More Like This" (term-overlap similarity) instead of vector
-    similarity - cruder, but far better than no candidates at all.
-    more_like_this_ids returns only a ranked ID list, no scores, so weight is
-    synthesized from rank (descending from top_k) rather than claiming a
-    similarity magnitude that doesn't exist. An unrestricted user (none, or an
-    active superuser - see user_is_unrestricted) is normalized to ``None``
-    before calling, since the backend's permission filter has no superuser
-    short-circuit of its own. Results are re-checked with
-    restrict_queryset_to_visible() since Tantivy's indexed permission fields
-    lag the DB via async reindexing.
-    """
-    from documents.search import get_backend
-
-    unrestricted = user_is_unrestricted(user)
-    search_user = None if unrestricted else user
-    backend = get_backend()
-    similar_ids = backend.more_like_this_ids(
-        document.pk,
-        user=search_user,
-        limit=top_k,
-    )
-    if not unrestricted:
-        allowed_ids = set(
-            restrict_queryset_to_visible(
-                Document.objects.filter(pk__in=similar_ids),
-                user,
-                "view_document",
-            ).values_list("pk", flat=True),
-        )
-        similar_ids = [doc_id for doc_id in similar_ids if doc_id in allowed_ids]
-    return [
-        SimilarDocument(document_id=doc_id, weight=float(top_k - rank))
-        for rank, doc_id in enumerate(similar_ids)
-    ]
-
-
 def get_language_name(language_code: str) -> str:
     normalized_language_code = language_code.lower()
     for code, name in settings.LANGUAGES:
@@ -181,52 +136,43 @@ def get_taxonomy_context(
     user: User | None = None,
     max_docs: int = 5,
 ) -> tuple[TaxonomyCandidates, str]:
-    """One retrieval feeds both taxonomy candidates and RAG text context. Uses
+    """One retrieval feeds both taxonomy candidates and RAG text context.
-    vector similarity when an embedding backend is configured, otherwise
+    On any retrieval failure, degrades to empty candidates/context rather than
-    falls back to Tantivy full-text "More Like This" similarity - see
+    propagating the exception - a vector-store outage should not block
-    _fulltext_similar_documents. On any retrieval failure, degrades to empty
+    classification, only its RAG-assisted enrichment.
-    candidates/context rather than propagating the exception - neither a
-    vector-store outage nor a search-index issue should block classification,
-    only its context-assisted enrichment.
     """
-    ai_config = AIConfig()
     try:
-        if ai_config.llm_embedding_backend:
+        # None means "no restriction" to retrieve_similar_nodes. A superuser
-            # None means "no restriction" to retrieve_similar_nodes. An
+        # (like no user at all) can see every document, so skip materializing
-            # unrestricted user (no user at all, or an active superuser -- see
+        # every visible pk into a Python list and passing it through as an IN
-            # user_is_unrestricted) can see every document, so skip
+        # filter: for a large library that is a wasted quadratic scan in the
-            # materializing every visible pk into a Python list and passing it
+        # vector store at best, and past ~32,763 documents a hard
-            # through as an IN filter: for a large library that is a wasted
+        # sqlite3.OperationalError (SQLite's bound-parameter limit) at worst.
-            # quadratic scan in the vector store at best, and past ~32,763
+        # get_objects_for_user_owner_aware() would return every Document for a
-            # documents a hard sqlite3.OperationalError (SQLite's
+        # superuser anyway (guardian's own with_superuser shortcut), so this
-            # bound-parameter limit) at worst.
+        # changes nothing about which documents are considered -- only how we
-            # permitted_object_ids() has its own superuser shortcut that would
+        # get there.
-            # return every Document's id anyway, so this changes nothing about
+        visible_document_ids = (
-            # which documents are considered -- only how we get there.
+            None
-            visible_document_ids = (
+            if user is None or user.is_superuser
-                None
+            else list(
-                if user_is_unrestricted(user)
+                get_objects_for_user_owner_aware(
-                else list(permitted_object_ids(user, Document, "view_document"))
+                    user,
-            )
+                    "view_document",
-            nodes = retrieve_similar_nodes(
+                    Document,
-                document,
+                ).values_list("pk", flat=True),
-                top_k=TAXONOMY_CANDIDATE_TOP_K,
-                document_ids=visible_document_ids,
-            )
-            similar_documents = _node_document_weights(nodes)
-        else:
-            # See _fulltext_similar_documents: it applies its own permission
-            # filter via `user`, so no visible-document-id list is needed here.
-            similar_documents = _fulltext_similar_documents(
-                document,
-                user,
-                top_k=TAXONOMY_CANDIDATE_TOP_K,
             )
+        )
+        nodes = retrieve_similar_nodes(
+            document,
+            top_k=TAXONOMY_CANDIDATE_TOP_K,
+            document_ids=visible_document_ids,
+        )
 
-        candidates = build_taxonomy_candidates(similar_documents, user)
+        candidates = build_taxonomy_candidates(nodes, user)
 
-        # similar_documents is already ordered by descending weight; don't lose it.
+        # ``nodes`` are already ordered by descending vector similarity; don't lose it.
-        similar_document_ids = [s["document_id"] for s in similar_documents]
+        similar_document_ids = list(dict.fromkeys(_node_document_ids(nodes)))
         similar_documents_by_id = Document.objects.in_bulk(similar_document_ids)
         similar_docs = [
             similar_documents_by_id[document_id]
@@ -240,8 +186,8 @@ def get_taxonomy_context(
             context_blocks.append(f"TITLE: {title}\n{text}")
     except Exception:
         logger.exception(
-            "Failed to retrieve similar-document context for document %s; "
+            "Failed to retrieve RAG neighbours for document %s; continuing "
-            "continuing without taxonomy candidates or similar-document context.",
+            "without taxonomy candidates or similar-document context.",
             document.pk,
         )
         return empty_taxonomy_candidates(), ""
@@ -295,13 +241,17 @@ def get_ai_document_classification(
 ) -> ClassificationSuggestions:
     ai_config = AIConfig()
 
-    candidates, context = get_taxonomy_context(document, user)
+    if ai_config.llm_embedding_backend:
-    prompt = build_prompt_with_rag(
+        candidates, context = get_taxonomy_context(document, user)
-        document,
+        prompt = build_prompt_with_rag(
-        ai_config,
+            document,
-        candidates=candidates,
+            ai_config,
-        context=context,
+            candidates=candidates,
-    )
+            context=context,
+        )
+    else:
+        candidates = empty_taxonomy_candidates()
+        prompt = build_prompt_without_rag(document, ai_config, candidates=candidates)
 
     client = AIClient()
     # Hand the pooled DB connection back while the (slow) LLM query runs so it

src/paperless_ai/client.py

@@ -22,7 +22,6 @@ from paperless.network import validate_outbound_http_url
 from paperless_ai.base_model import ClassificationSuggestions
 from paperless_ai.base_model import DocumentClassifierSchema
 from paperless_ai.base_model import model_to_classification_suggestions
-from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 
 logger = logging.getLogger("paperless_ai.client")
@@ -133,7 +132,7 @@ class AIClient:
         from llama_index.core.llms import ChatMessage
 
         if self.settings.llm_backend == LLMBackend.OLLAMA:
-            with self._normalize_errors():
+            with self._normalize_timeouts():
                 result = self.llm.chat(
                     [ChatMessage(role="user", content=prompt)],
                     format=DocumentClassifierSchema.model_json_schema(),
@@ -154,7 +153,7 @@ class AIClient:
             content=f"{prompt}\n\n"
             f"Answer by calling the {tool.metadata.name} tool. Do not write the answer as text.",
         )
-        with self._normalize_errors():
+        with self._normalize_timeouts():
             result = self.llm.chat_with_tools(
                 tools=[tool],
                 user_msg=user_msg,
@@ -174,7 +173,7 @@ class AIClient:
         )
 
     @contextmanager
-    def _normalize_errors(self) -> Iterator[None]:
+    def _normalize_timeouts(self) -> Iterator[None]:
         try:
             yield
         except httpx.TimeoutException as exc:
@@ -182,23 +181,8 @@ class AIClient:
         except Exception as exc:
             if self._is_openai_timeout(exc):
                 raise LLMTimeoutError from exc
-            if self._is_provider_error(exc):
-                raise LLMProviderError from exc
             raise
 
-    def _is_provider_error(self, exc: Exception) -> bool:
-        if self.settings.llm_backend == LLMBackend.OLLAMA:
-            from ollama import ResponseError
-
-            return isinstance(exc, ResponseError)
-
-        if self.settings.llm_backend == LLMBackend.OPENAI_LIKE:
-            from openai import APIStatusError
-
-            return isinstance(exc, APIStatusError)
-
-        return False
-
     def _is_openai_timeout(self, exc: Exception) -> bool:
         if self.settings.llm_backend != LLMBackend.OPENAI_LIKE:
             return False

src/paperless_ai/exceptions.py

@@ -1,6 +1,2 @@
 class LLMTimeoutError(Exception):
     pass
-
-
-class LLMProviderError(Exception):
-    """The LLM backend rejected the request."""

src/paperless_ai/indexing.py

@@ -721,3 +721,20 @@ def retrieve_similar_nodes(
             continue
         filtered.append(node)
     return filtered
+
+
+def _node_document_ids(nodes: list["NodeWithScore"]) -> list[int]:
+    document_ids: list[int] = []
+    for node in nodes:
+        document_id = node.metadata.get("document_id")
+        if document_id is None:  # pragma: no cover
+            # See the matching guard in retrieve_similar_nodes() above.
+            continue
+        try:
+            document_ids.append(int(document_id))
+        except ValueError:  # pragma: no cover
+            logger.warning(
+                "Skipping LLM index result with invalid document_id %r.",
+                document_id,
+            )
+    return document_ids

src/paperless_ai/taxonomy.py

@@ -31,11 +31,6 @@ class TaxonomyCandidate(TypedDict):
     weight: float
 
 
-class SimilarDocument(TypedDict):
-    document_id: int
-    weight: float
-
-
 class TaxonomyCandidates(TypedDict):
     tags: list[TaxonomyCandidate]
     document_types: list[TaxonomyCandidate]
@@ -54,10 +49,10 @@ def empty_taxonomy_candidates() -> TaxonomyCandidates:
     )
 
 
-def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument]:
+def _node_document_weights(nodes: list["NodeWithScore"]) -> dict[int, float]:
-    """Sum each node's similarity score into its document_id (a document can
+    """document_id -> that node's similarity score, summed if a document_id
-    appear via multiple chunks/nodes) and return one SimilarDocument per
+    appears more than once across the retrieved nodes (e.g. multiple chunks
-    distinct document_id."""
+    of the same source document)."""
     weights: dict[int, float] = defaultdict(float)
     for node in nodes:
         document_id = node.metadata.get("document_id")
@@ -70,14 +65,7 @@ def _node_document_weights(nodes: list["NodeWithScore"]) -> list[SimilarDocument
             weights[int(document_id)] += float(node.score or 0.0)
         except (TypeError, ValueError):  # pragma: no cover
             continue
-    return sorted(
+    return weights
-        (
-            SimilarDocument(document_id=document_id, weight=weight)
-            for document_id, weight in weights.items()
-        ),
-        key=lambda similar: similar["weight"],
-        reverse=True,
-    )
 
 
 def _visible_ranked_candidates(
@@ -113,25 +101,20 @@ def _visible_ranked_candidates(
 
 
 def build_taxonomy_candidates(
-    similar_documents: list[SimilarDocument],
+    nodes: list["NodeWithScore"],
     user: User | None,
 ) -> TaxonomyCandidates:
-    """Resolve each similar document's id to a live Document, read its
+    """Resolve each neighbour node's document_id to a live Document, read its
-    *current* tags/type/correspondent/storage_path via the ORM (never any
+    *current* tags/type/correspondent/storage_path via the ORM (never the
-    possibly-stale names an adapter's source might have cached), weight each
+    possibly-stale names cached in vector-index node metadata), weight each
-    distinct taxonomy object by aggregate similarity weight, permission-filter
+    distinct taxonomy object by aggregate neighbour similarity, permission-filter
     against what ``user`` can see, and return each category ranked by weight
-    and capped. ``similar_documents`` may come from either the vector-RAG
+    and capped.
-    adapter or the full-text fallback adapter - both produce this same shape.
     """
-    if not similar_documents:
-        return empty_taxonomy_candidates()
 
-    # Both adapters guarantee at most one SimilarDocument per document_id, so
+    document_weights = _node_document_weights(nodes)
-    # this never silently drops a duplicate's weight.
+    if not document_weights:
-    document_weights: dict[int, float] = {
+        return empty_taxonomy_candidates()
-        s["document_id"]: s["weight"] for s in similar_documents
-    }
 
     # Only .tags.all() needs prefetching (a reverse M2M, one extra query for
     # the whole batch). document_type/correspondent/storage_path are read

src/paperless_ai/tests/test_ai_classifier.py

@@ -1,5 +1,4 @@
 import datetime
-from collections.abc import Generator
 from types import SimpleNamespace
 from unittest.mock import MagicMock
 from unittest.mock import patch
@@ -7,24 +6,18 @@ from unittest.mock import patch
 import pytest
 import pytest_mock
 from django.test import override_settings
-from guardian.shortcuts import assign_perm
-from guardian.shortcuts import remove_perm
 
 from documents.models import Document
-from documents.search import TantivyBackend
 from documents.tests.factories import DocumentFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
 from paperless.config import AIConfig
-from paperless_ai.ai_classifier import TAXONOMY_CANDIDATE_TOP_K
-from paperless_ai.ai_classifier import _fulltext_similar_documents
 from paperless_ai.ai_classifier import build_localization_prompt
 from paperless_ai.ai_classifier import build_prompt_with_rag
 from paperless_ai.ai_classifier import build_prompt_without_rag
 from paperless_ai.ai_classifier import get_ai_document_classification
 from paperless_ai.ai_classifier import get_language_name
 from paperless_ai.ai_classifier import get_taxonomy_context
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidate
 from paperless_ai.taxonomy import TaxonomyCandidates
 
@@ -227,10 +220,12 @@ def test_use_rag_if_configured(
 
 @pytest.mark.django_db
 @patch("paperless_ai.client.AIClient.run_llm_query")
-@patch("paperless_ai.ai_classifier.build_prompt_with_rag")
+@patch("paperless_ai.ai_classifier.build_prompt_without_rag")
+@patch("paperless_ai.ai_classifier.AIConfig")
 @override_settings(LLM_BACKEND="ollama", LLM_MODEL="some_model")
-def test_use_rag_prompt_even_without_embedding_backend(
+def test_use_without_rag_if_not_configured(
-    mock_build_prompt_with_rag,
+    mock_ai_config,
+    mock_build_prompt_without_rag,
     mock_run_llm_query,
     mock_document,
 ):
@@ -240,13 +235,13 @@ def test_use_rag_prompt_even_without_embedding_backend(
     WHEN:
         - get_ai_document_classification() is called
     THEN:
-        - The RAG-context prompt builder is still used (fed by the full-text
+        - The non-RAG prompt builder is used
-          fallback's context/candidates instead of the vector store's)
     """
-    mock_build_prompt_with_rag.return_value = "Prompt with RAG"
+    mock_ai_config.return_value.llm_embedding_backend = None
+    mock_build_prompt_without_rag.return_value = "Prompt without RAG"
     mock_run_llm_query.return_value = NESTED_SUGGESTIONS
     get_ai_document_classification(mock_document)
-    mock_build_prompt_with_rag.assert_called_once()
+    mock_build_prompt_without_rag.assert_called_once()
 
 
 @pytest.mark.django_db
@@ -325,7 +320,6 @@ def test_build_localization_prompt_preserves_unicode_characters():
 
 
 @pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_assembles_rag_text_and_candidates():
     """
     GIVEN:
@@ -360,7 +354,6 @@ def test_get_taxonomy_context_assembles_rag_text_and_candidates():
 
 
 @pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_preserves_similarity_order_and_distinct_documents():
     """
     GIVEN:
@@ -431,7 +424,6 @@ def test_get_taxonomy_context_preserves_similarity_order_and_distinct_documents(
 
 
 @pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 def test_get_taxonomy_context_no_similar_docs():
     """
     GIVEN:
@@ -455,67 +447,6 @@ def test_get_taxonomy_context_no_similar_docs():
     }
 
 
-@pytest.mark.django_db
-def test_get_taxonomy_context_uses_fulltext_fallback_when_no_embedding_backend(
-    mocker: pytest_mock.MockerFixture,
-) -> None:
-    """
-    GIVEN:
-        - No LLM embedding backend is configured (the default test settings)
-    WHEN:
-        - get_taxonomy_context() is called
-    THEN:
-        - _fulltext_similar_documents() is called with the document, the user
-          and TAXONOMY_CANDIDATE_TOP_K
-        - retrieve_similar_nodes() (the vector path) is never called
-    """
-    document = DocumentFactory.create(content="Some content")
-    mock_fulltext = mocker.patch(
-        "paperless_ai.ai_classifier._fulltext_similar_documents",
-        return_value=[],
-    )
-    mock_retrieve = mocker.patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
-
-    get_taxonomy_context(document, user=None)
-
-    mock_fulltext.assert_called_once_with(
-        document,
-        None,
-        top_k=TAXONOMY_CANDIDATE_TOP_K,
-    )
-    mock_retrieve.assert_not_called()
-
-
-@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
-def test_get_taxonomy_context_uses_vector_path_when_embedding_backend_configured(
-    mocker: pytest_mock.MockerFixture,
-) -> None:
-    """
-    GIVEN:
-        - An LLM embedding backend is configured
-    WHEN:
-        - get_taxonomy_context() is called
-    THEN:
-        - retrieve_similar_nodes() (the vector path) is called
-        - _fulltext_similar_documents() (the no-embedding-backend fallback)
-          is never called
-    """
-    document = DocumentFactory.create(content="Some content")
-    mock_retrieve = mocker.patch(
-        "paperless_ai.ai_classifier.retrieve_similar_nodes",
-        return_value=[],
-    )
-    mock_fulltext = mocker.patch(
-        "paperless_ai.ai_classifier._fulltext_similar_documents",
-    )
-
-    get_taxonomy_context(document, user=None)
-
-    mock_retrieve.assert_called_once()
-    mock_fulltext.assert_not_called()
-
-
 class TestGetTaxonomyContextVisibility:
     """get_taxonomy_context must not materialize every visible document id
     for a user who can already see the whole library: a superuser (like no
@@ -528,7 +459,6 @@ class TestGetTaxonomyContextVisibility:
     """
 
     @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_skips_permission_lookup_for_superuser(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -547,18 +477,17 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_permitted = mocker.patch(
+        mock_get_objects = mocker.patch(
-            "paperless_ai.ai_classifier.permitted_object_ids",
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
         )
         user = UserFactory.create(is_superuser=True)
 
         get_taxonomy_context(document, user)
 
-        mock_permitted.assert_not_called()
+        mock_get_objects.assert_not_called()
         assert mock_retrieve.call_args.kwargs["document_ids"] is None
 
     @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_skips_permission_lookup_when_no_user(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -577,17 +506,16 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_permitted = mocker.patch(
+        mock_get_objects = mocker.patch(
-            "paperless_ai.ai_classifier.permitted_object_ids",
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
         )
 
         get_taxonomy_context(document, None)
 
-        mock_permitted.assert_not_called()
+        mock_get_objects.assert_not_called()
         assert mock_retrieve.call_args.kwargs["document_ids"] is None
 
     @pytest.mark.django_db
-    @override_settings(LLM_EMBEDDING_BACKEND="huggingface")
     def test_restricts_to_visible_documents_for_non_superuser(
         self,
         mocker: pytest_mock.MockerFixture,
@@ -598,7 +526,7 @@ class TestGetTaxonomyContextVisibility:
         WHEN:
             - get_taxonomy_context() is called
         THEN:
-            - The user's permitted document ids are looked up and passed to
+            - The user's visible document ids are looked up and passed to
               retrieve_similar_nodes() as a restriction
         """
         document = DocumentFactory.create(content="Some content")
@@ -606,232 +534,21 @@ class TestGetTaxonomyContextVisibility:
             "paperless_ai.ai_classifier.retrieve_similar_nodes",
             return_value=[],
         )
-        mock_permitted = mocker.patch(
+        mock_queryset = mocker.MagicMock()
-            "paperless_ai.ai_classifier.permitted_object_ids",
+        mock_queryset.values_list.return_value = [1, 2, 3]
-            return_value=[1, 2, 3],
+        mock_get_objects = mocker.patch(
+            "paperless_ai.ai_classifier.get_objects_for_user_owner_aware",
+            return_value=mock_queryset,
         )
         user = UserFactory.create(is_superuser=False)
 
         get_taxonomy_context(document, user)
 
-        mock_permitted.assert_called_once_with(user, Document, "view_document")
+        mock_get_objects.assert_called_once_with(user, "view_document", Document)
         assert mock_retrieve.call_args.kwargs["document_ids"] == [1, 2, 3]
 
 
 @pytest.mark.django_db
-class TestFulltextSimilarDocuments:
-    """_fulltext_similar_documents is the no-embedding-backend fallback: it
-    asks the Tantivy full-text index for "More Like This" neighbours instead
-    of the vector store, and synthesizes a rank-based weight since Tantivy's
-    more_like_this_ids returns only an ordered id list, no scores.
-    """
-
-    @pytest.fixture
-    def fulltext_backend(
-        self,
-        mocker: pytest_mock.MockerFixture,
-    ) -> Generator[TantivyBackend, None, None]:
-        """An in-memory Tantivy backend, wired up as the module-level
-        singleton _fulltext_similar_documents resolves via get_backend()."""
-        backend = TantivyBackend(path=None)
-        backend.open()
-        mocker.patch("documents.search.get_backend", return_value=backend)
-        try:
-            yield backend
-        finally:
-            backend.close()
-
-    def test_ranks_by_rank_based_weight_descending(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and two similar documents indexed in Tantivy
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - Each result's weight reflects its rank (first result weighted
-              higher than the second), not a raw similarity score
-        """
-        source = DocumentFactory.create(content="quarterly financial report details")
-        first = DocumentFactory.create(content="quarterly financial report details")
-        second = DocumentFactory.create(content="financial report")
-        for doc in (source, first, second):
-            fulltext_backend.add_or_update(doc)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert len(result) == 2
-        weight_by_id = {s["document_id"]: s["weight"] for s in result}
-        assert weight_by_id[first.pk] > weight_by_id[second.pk]
-
-    def test_excludes_source_document(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document indexed in Tantivy with no other documents
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - An empty list is returned - the source document is never its
-              own similar document
-        """
-        source = DocumentFactory.create(content="unique unrelated content")
-        fulltext_backend.add_or_update(source)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert result == []
-
-    def test_empty_index_returns_empty_list(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A document that has never been indexed (fresh/empty Tantivy index)
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - An empty list is returned rather than raising
-        """
-        source = DocumentFactory.create(content="never indexed")
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        assert result == []
-
-    def test_respects_top_k_limit(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and four similar documents indexed
-        WHEN:
-            - _fulltext_similar_documents() is called with top_k=2
-        THEN:
-            - At most 2 results are returned
-        """
-        source = DocumentFactory.create(content="shared overlapping keyword text")
-        fulltext_backend.add_or_update(source)
-        for _ in range(4):
-            fulltext_backend.add_or_update(
-                DocumentFactory.create(content="shared overlapping keyword text"),
-            )
-
-        result = _fulltext_similar_documents(source, user=None, top_k=2)
-
-        assert len(result) == 2
-
-    def test_result_shape_is_similar_document(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document and one similar document indexed
-        WHEN:
-            - _fulltext_similar_documents() is called
-        THEN:
-            - Each result is a SimilarDocument (document_id + weight only)
-        """
-        source = DocumentFactory.create(content="shared content phrase")
-        other = DocumentFactory.create(content="shared content phrase")
-        fulltext_backend.add_or_update(source)
-        fulltext_backend.add_or_update(other)
-
-        result = _fulltext_similar_documents(source, user=None, top_k=5)
-
-        # rank 0 (the only/best result) with top_k=5 -> weight = top_k - rank = 5.0,
-        # per the "first result gets top_k, the last gets 1" formula.
-        assert result == [SimilarDocument(document_id=other.pk, weight=5.0)]
-
-    def test_superuser_sees_other_users_documents(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A source document owned by one user and a similar document
-              owned by a different user, with no sharing between them
-        WHEN:
-            - _fulltext_similar_documents() is called with a superuser
-        THEN:
-            - The other user's document is still returned as a similar
-              document - a superuser must not be narrowed by the backend's
-              owner-based permission filter
-        """
-        owner = UserFactory.create()
-        other_owner = UserFactory.create()
-        superuser = UserFactory.create(is_superuser=True)
-        source = DocumentFactory.create(
-            content="shared content phrase",
-            owner=owner,
-        )
-        other = DocumentFactory.create(
-            content="shared content phrase",
-            owner=other_owner,
-        )
-        fulltext_backend.add_or_update(source)
-        fulltext_backend.add_or_update(other)
-
-        result = _fulltext_similar_documents(source, user=superuser, top_k=5)
-
-        assert [s["document_id"] for s in result] == [other.pk]
-
-    def test_excludes_stale_permitted_document_for_regular_user(
-        self,
-        fulltext_backend: TantivyBackend,
-    ) -> None:
-        """
-        GIVEN:
-            - A regular (non-superuser) user
-            - A similar document the user is permitted to view, and another
-              similar document indexed while the user still had view
-              permission but which has since had that permission revoked in
-              the database, i.e. the Tantivy index has stale permission data
-        WHEN:
-            - _fulltext_similar_documents() is called with that user
-        THEN:
-            - Only the still-permitted document is returned - the DB
-              re-check via restrict_queryset_to_visible() must catch the
-              document Tantivy's stale index still thinks is visible
-        """
-        owner = UserFactory.create()
-        viewer = UserFactory.create(is_superuser=False)
-        source = DocumentFactory.create(
-            content="shared content phrase",
-            owner=owner,
-        )
-        permitted = DocumentFactory.create(
-            content="shared content phrase",
-            owner=owner,
-        )
-        now_private = DocumentFactory.create(
-            content="shared content phrase",
-            owner=owner,
-        )
-        assign_perm("view_document", viewer, permitted)
-        assign_perm("view_document", viewer, now_private)
-        fulltext_backend.add_or_update(source)
-        fulltext_backend.add_or_update(permitted)
-        fulltext_backend.add_or_update(now_private)
-
-        # Revoke access after indexing, without reindexing: the index still
-        # carries viewer as a permitted viewer for `now_private`.
-        remove_perm("view_document", viewer, now_private)
-
-        result = _fulltext_similar_documents(source, user=viewer, top_k=5)
-
-        assert [s["document_id"] for s in result] == [permitted.pk]
-
-
-@pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 @patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrieve):
     """
@@ -858,7 +575,6 @@ def test_get_taxonomy_context_retrieval_failure_degrades_to_no_hints(mock_retrie
 
 
 @pytest.mark.django_db
-@override_settings(LLM_EMBEDDING_BACKEND="huggingface")
 @patch("paperless_ai.ai_classifier.build_taxonomy_candidates")
 @patch("paperless_ai.ai_classifier.retrieve_similar_nodes")
 def test_get_taxonomy_context_candidate_building_failure_degrades_to_no_hints(

src/paperless_ai/tests/test_ai_indexing.py

@@ -1188,7 +1188,9 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, document_ids=[b.id])
 
-        assert all(int(node.metadata["document_id"]) == b.id for node in nodes)
+        assert all(
+            document_id == b.id for document_id in indexing._node_document_ids(nodes)
+        )
 
     def test_excludes_self(
         self,
@@ -1210,7 +1212,7 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, top_k=5)
 
-        assert {int(node.metadata["document_id"]) for node in nodes} == {b.id}
+        assert set(indexing._node_document_ids(nodes)) == {b.id}
 
     def test_excludes_self_with_multiple_chunks(
         self,
@@ -1233,4 +1235,4 @@ class TestRetrieveSimilarNodesAgainstRealIndex:
 
         nodes = indexing.retrieve_similar_nodes(a, top_k=3)
 
-        assert {int(node.metadata["document_id"]) for node in nodes} == {b.id}
+        assert set(indexing._node_document_ids(nodes)) == {b.id}

src/paperless_ai/tests/test_client.py

@@ -4,7 +4,6 @@ from unittest.mock import MagicMock
 from unittest.mock import patch
 
 import httpx
-import ollama
 import openai
 import pytest
 from llama_index.core.llms.llm import ToolSelection
@@ -12,7 +11,6 @@ from llama_index.core.llms.llm import ToolSelection
 from paperless_ai.client import LLM_SYSTEM_PROMPT
 from paperless_ai.client import PLACEHOLDER_API_KEY
 from paperless_ai.client import AIClient
-from paperless_ai.exceptions import LLMProviderError
 from paperless_ai.exceptions import LLMTimeoutError
 
 
@@ -216,52 +214,6 @@ def test_run_llm_query_openai_timeout_raises_local_error(
         client.run_llm_query("test_prompt")
 
 
-def test_run_llm_query_openai_status_error_raises_provider_error(
-    mock_ai_config,
-    mock_openai_llm,
-):
-    mock_ai_config.llm_backend = "openai-like"
-    mock_ai_config.llm_model = "test_model"
-    mock_ai_config.llm_endpoint = "http://test-url"
-
-    request = httpx.Request("POST", "http://test-url/v1/chat/completions")
-    body = {"error": {"message": "Thinking mode does not support this tool_choice"}}
-    mock_openai_llm.return_value.chat_with_tools.side_effect = openai.BadRequestError(
-        "Error code: 400",
-        response=httpx.Response(400, request=request, json=body),
-        body=body,
-    )
-
-    client = AIClient()
-
-    with pytest.raises(LLMProviderError) as exc_info:
-        client.run_llm_query("test_prompt")
-    assert str(exc_info.value) == ""
-    assert isinstance(exc_info.value.__cause__, openai.BadRequestError)
-
-
-def test_run_llm_query_ollama_response_error_raises_provider_error(
-    mock_ai_config,
-    mock_ollama_llm,
-):
-    mock_ai_config.llm_backend = "ollama"
-    mock_ai_config.llm_model = "test_model"
-    mock_ai_config.llm_endpoint = "http://test-url"
-
-    response_error = ollama.ResponseError(
-        "confidential provider response",
-        status_code=400,
-    )
-    mock_ollama_llm.return_value.chat.side_effect = response_error
-
-    client = AIClient()
-
-    with pytest.raises(LLMProviderError) as exc_info:
-        client.run_llm_query("test_prompt")
-    assert str(exc_info.value) == ""
-    assert exc_info.value.__cause__ is response_error
-
-
 def test_run_llm_query_httpx_timeout_raises_local_error(
     mock_ai_config,
     mock_ollama_llm,

src/paperless_ai/tests/test_taxonomy.py

@@ -1,4 +1,5 @@
 import json
+from types import SimpleNamespace
 
 import pytest
 import pytest_mock
@@ -9,14 +10,14 @@ from documents.tests.factories import DocumentTypeFactory
 from documents.tests.factories import StoragePathFactory
 from documents.tests.factories import TagFactory
 from documents.tests.factories import UserFactory
-from paperless_ai.taxonomy import SimilarDocument
 from paperless_ai.taxonomy import TaxonomyCandidates
 from paperless_ai.taxonomy import build_taxonomy_candidates
 from paperless_ai.taxonomy import format_taxonomy_for_prompt
 
 
-def make_similar(document_id: int, weight: float) -> SimilarDocument:
+def make_node(document_id: int, score: float) -> SimpleNamespace:
-    return SimilarDocument(document_id=document_id, weight=weight)
+    """A stand-in for NodeWithScore: only ``.metadata``/``.score`` are read."""
+    return SimpleNamespace(metadata={"document_id": str(document_id)}, score=score)
 
 
 @pytest.mark.django_db
@@ -52,9 +53,9 @@ class TestBuildTaxonomyCandidates:
         doc_a.tags.add(tag)
         doc_b = DocumentFactory.create()
         doc_b.tags.add(tag)
-        similar_documents = [make_similar(doc_a.pk, 0.9), make_similar(doc_b.pk, 0.4)]
+        nodes = [make_node(doc_a.pk, 0.9), make_node(doc_b.pk, 0.4)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["tags"]) == 1
         assert result["tags"][0]["id"] == tag.pk
@@ -79,9 +80,9 @@ class TestBuildTaxonomyCandidates:
         document.tags.add(tag)
         tag.name = "New Name"
         tag.save()
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert result["tags"][0]["name"] == "New Name"
 
@@ -101,9 +102,9 @@ class TestBuildTaxonomyCandidates:
         document = DocumentFactory.create()
         document.tags.add(tag)
         tag.delete()
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert result["tags"] == []
 
@@ -122,12 +123,9 @@ class TestBuildTaxonomyCandidates:
         strong_doc.tags.add(strong_tag)
         weak_doc = DocumentFactory.create()
         weak_doc.tags.add(weak_tag)
-        similar_documents = [
+        nodes = [make_node(strong_doc.pk, 0.9), make_node(weak_doc.pk, 0.1)]
-            make_similar(strong_doc.pk, 0.9),
-            make_similar(weak_doc.pk, 0.1),
-        ]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert [c["name"] for c in result["tags"]] == ["Strong", "Weak"]
 
@@ -143,9 +141,9 @@ class TestBuildTaxonomyCandidates:
         document = DocumentFactory.create()
         for i in range(15):
             document.tags.add(TagFactory.create(name=f"Tag{i}"))
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["tags"]) == 10
 
@@ -159,12 +157,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 correspondents are returned
         """
         correspondents = CorrespondentFactory.create_batch(7)
-        similar_documents = [
+        nodes = [
-            make_similar(DocumentFactory.create(correspondent=c).pk, 0.5)
+            make_node(DocumentFactory.create(correspondent=c).pk, 0.5)
             for c in correspondents
         ]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["correspondents"]) == 5
 
@@ -179,9 +177,9 @@ class TestBuildTaxonomyCandidates:
         """
         document_type = DocumentTypeFactory.create(name="Invoice")
         document = DocumentFactory.create(document_type=document_type)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["document_types"]) == 1
         assert result["document_types"][0]["id"] == document_type.pk
@@ -197,12 +195,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 document_types are returned
         """
         document_types = DocumentTypeFactory.create_batch(7)
-        similar_documents = [
+        nodes = [
-            make_similar(DocumentFactory.create(document_type=dt).pk, 0.5)
+            make_node(DocumentFactory.create(document_type=dt).pk, 0.5)
             for dt in document_types
         ]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["document_types"]) == 5
 
@@ -217,9 +215,9 @@ class TestBuildTaxonomyCandidates:
         """
         storage_path = StoragePathFactory.create(name="Invoices")
         document = DocumentFactory.create(storage_path=storage_path)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["storage_paths"]) == 1
         assert result["storage_paths"][0]["id"] == storage_path.pk
@@ -235,12 +233,12 @@ class TestBuildTaxonomyCandidates:
             - Only 5 storage_paths are returned
         """
         storage_paths = StoragePathFactory.create_batch(7)
-        similar_documents = [
+        nodes = [
-            make_similar(DocumentFactory.create(storage_path=sp).pk, 0.5)
+            make_node(DocumentFactory.create(storage_path=sp).pk, 0.5)
             for sp in storage_paths
         ]
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert len(result["storage_paths"]) == 5
 
@@ -260,14 +258,14 @@ class TestBuildTaxonomyCandidates:
         tag = TagFactory.create(name="Restricted")
         document = DocumentFactory.create()
         document.tags.add(tag)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
         user = UserFactory.create()
         mocker.patch(
             "documents.permissions.permitted_object_ids",
             return_value=[],  # user cannot see this tag
         )
 
-        result = build_taxonomy_candidates(similar_documents, user=user)
+        result = build_taxonomy_candidates(nodes, user=user)
 
         assert result["tags"] == []
 
@@ -297,10 +295,10 @@ class TestBuildTaxonomyCandidates:
         tag.save()
         document = DocumentFactory.create()
         document.tags.add(tag)
-        similar_documents = [make_similar(document.pk, 0.5)]
+        nodes = [make_node(document.pk, 0.5)]
         spy = mocker.patch("documents.permissions.permitted_object_ids")
 
-        result = build_taxonomy_candidates(similar_documents, user=None)
+        result = build_taxonomy_candidates(nodes, user=None)
 
         assert result["tags"][0]["name"] == "Owned"
         spy.assert_not_called()

src/paperless_mail/mail.py

@@ -6,7 +6,6 @@ import socket
 import ssl
 import tempfile
 import traceback
-import unicodedata
 from datetime import date
 from datetime import timedelta
 from fnmatch import fnmatch
@@ -45,6 +44,7 @@ from documents.models import Correspondent
 from documents.models import PaperlessTask
 from documents.parsers import is_mime_type_supported
 from documents.tasks import consume_file
+from documents.utils import normalize_unicode
 from paperless.network import is_public_ip
 from paperless.network import resolve_hostname_ips
 from paperless_mail.models import MailAccount
@@ -617,10 +617,10 @@ class MailAccountHandler(LoggingMixin):
         rule: MailRule,
     ) -> str | None:
         if rule.assign_title_from == MailRule.TitleSource.FROM_SUBJECT:
-            return unicodedata.normalize("NFC", message.subject)
+            return normalize_unicode(message.subject)
 
         elif rule.assign_title_from == MailRule.TitleSource.FROM_FILENAME:
-            return unicodedata.normalize("NFC", Path(att.filename).stem)
+            return normalize_unicode(Path(att.filename).stem)
 
         elif rule.assign_title_from == MailRule.TitleSource.NONE:
             return None
@@ -1004,6 +1004,8 @@ class MailAccountHandler(LoggingMixin):
         consume_tasks = []
 
         for att in message.attachments:
+            attachment_filename = normalize_unicode(att.filename)
+
             if (
                 att.content_disposition != "attachment"
                 and rule.attachment_type
@@ -1018,7 +1020,7 @@ class MailAccountHandler(LoggingMixin):
 
             if not self.filename_inclusion_matches(
                 rule.filter_attachment_filename_include,
-                att.filename,
+                attachment_filename,
             ):
                 # Force the filename and pattern to the lowercase
                 # as this is system dependent otherwise
@@ -1030,7 +1032,7 @@ class MailAccountHandler(LoggingMixin):
                 continue
             elif self.filename_exclusion_matches(
                 rule.filter_attachment_filename_exclude,
-                att.filename,
+                attachment_filename,
             ):
                 self.log.debug(
                     f"Rule {rule}: "
@@ -1064,7 +1066,7 @@ class MailAccountHandler(LoggingMixin):
                 )
 
                 attachment_name = pathvalidate.sanitize_filename(
-                    unicodedata.normalize("NFC", att.filename),
+                    attachment_filename,
                 )
                 if attachment_name:
                     temp_filename = temp_dir / attachment_name
@@ -1175,7 +1177,7 @@ class MailAccountHandler(LoggingMixin):
         doc_overrides = DocumentMetadataOverrides(
             title=message.subject,
             filename=pathvalidate.sanitize_filename(
-                unicodedata.normalize("NFC", f"{message.subject}.eml"),
+                normalize_unicode(f"{message.subject}.eml"),
             ),
             correspondent_id=correspondent.id if correspondent else None,
             document_type_id=doc_type.id if doc_type else None,

src/paperless_mail/serialisers.py

@@ -8,6 +8,7 @@ from documents.serialisers import CorrespondentField
 from documents.serialisers import DocumentTypeField
 from documents.serialisers import OwnedObjectSerializer
 from documents.serialisers import TagsField
+from documents.utils import normalize_unicode
 from paperless_mail.models import MailAccount
 from paperless_mail.models import MailRule
 from paperless_mail.models import ProcessedMail
@@ -161,6 +162,12 @@ class MailRuleSerializer(OwnedObjectSerializer):
             raise serializers.ValidationError("Maximum mail age is unreasonably large.")
         return value
 
+    def validate_filter_attachment_filename_include(self, value):
+        return normalize_unicode(value)
+
+    def validate_filter_attachment_filename_exclude(self, value):
+        return normalize_unicode(value)
+
 
 class ProcessedMailSerializer(OwnedObjectSerializer):
     class Meta:

src/paperless_mail/tasks.py

@@ -1,9 +1,7 @@
 import logging
 
-from celery import Task
 from celery import shared_task
 
-from documents.models import PaperlessTask
 from paperless_mail.mail import MailAccountHandler
 from paperless_mail.mail import MailError
 from paperless_mail.models import MailAccount
@@ -12,26 +10,8 @@ from paperless_mail.models import MailRule
 logger = logging.getLogger("paperless.mail.tasks")
 
 
-@shared_task(bind=True)
+@shared_task
-def process_mail_accounts(self: Task, account_ids: list[int] | None = None) -> str:
+def process_mail_accounts(account_ids: list[int] | None = None) -> str:
-    # A scheduled check can still be running (or queued) when the next one
-    # ProcessedMail dedup only records a message once its
-    # handling has finished, so an overlapping run can still pick up the same
-    # not-yet-recorded message. Skip outright rather than race it.
-    other_mail_fetch_running = (
-        PaperlessTask.objects.filter(
-            task_type=PaperlessTask.TaskType.MAIL_FETCH,
-            status__in=[PaperlessTask.Status.PENDING, PaperlessTask.Status.STARTED],
-        )
-        .exclude(task_id=self.request.id)
-        .exists()
-    )
-    if other_mail_fetch_running:
-        logger.info(
-            "Mail account processing is already running; skipping this run.",
-        )
-        return "Skipped: mail account processing already in progress."
-
     total_new_documents = 0
     accounts = (
         MailAccount.objects.filter(pk__in=account_ids)

src/paperless_mail/tests/test_process_mail_accounts_overlap.py

@@ -1,134 +0,0 @@
-from typing import Final
-
-import pytest
-import pytest_mock
-
-from documents.models import PaperlessTask
-from documents.tests.factories import PaperlessTaskFactory
-from paperless_mail import tasks
-from paperless_mail.tests.factories import MailAccountFactory
-from paperless_mail.tests.factories import MailRuleFactory
-
-NO_DOCUMENTS_ADDED: Final = "No new documents were added."
-SKIPPED: Final = "Skipped: mail account processing already in progress."
-
-
-@pytest.mark.django_db
-@pytest.mark.usefixtures("account_with_rule")
-class TestProcessMailAccountsOverlap:
-    @pytest.fixture
-    def account_with_rule(self) -> None:
-        """An enabled mail account with a single enabled rule."""
-        account = MailAccountFactory.create()
-        MailRuleFactory.create(account=account, enabled=True)
-
-    @pytest.mark.parametrize(
-        ("status", "expected_result", "expected_call_count"),
-        [
-            pytest.param(
-                PaperlessTask.Status.PENDING,
-                SKIPPED,
-                0,
-                id="pending-task-blocks",
-            ),
-            pytest.param(
-                PaperlessTask.Status.STARTED,
-                SKIPPED,
-                0,
-                id="started-task-blocks",
-            ),
-            pytest.param(
-                PaperlessTask.Status.SUCCESS,
-                NO_DOCUMENTS_ADDED,
-                1,
-                id="finished-task-does-not-block",
-            ),
-        ],
-    )
-    def test_skips_only_while_another_mail_fetch_task_runs(
-        self,
-        mocker: pytest_mock.MockerFixture,
-        status: PaperlessTask.Status,
-        expected_result: str,
-        expected_call_count: int,
-    ) -> None:
-        """
-        GIVEN:
-            - An enabled mail account with a rule
-            - Another mail fetch task row in the given status
-        WHEN:
-            - Mail accounts are processed
-        THEN:
-            - Processing is skipped only if that other task is pending or running
-        """
-        PaperlessTaskFactory.create(
-            task_type=PaperlessTask.TaskType.MAIL_FETCH,
-            trigger_source=PaperlessTask.TriggerSource.SCHEDULED,
-            status=status,
-        )
-
-        mocked_handle = mocker.patch.object(
-            tasks.MailAccountHandler,
-            "handle_mail_account",
-            return_value=0,
-        )
-
-        result = tasks.process_mail_accounts()
-
-        assert mocked_handle.call_count == expected_call_count
-        assert result == expected_result
-
-    def test_runs_when_no_other_mail_fetch_task_exists(
-        self,
-        mocker: pytest_mock.MockerFixture,
-    ) -> None:
-        """
-        GIVEN:
-            - An enabled mail account with a rule
-            - No other mail fetch task rows
-        WHEN:
-            - Mail accounts are processed
-        THEN:
-            - The account is handled
-        """
-        mocked_handle = mocker.patch.object(
-            tasks.MailAccountHandler,
-            "handle_mail_account",
-            return_value=0,
-        )
-
-        result = tasks.process_mail_accounts()
-
-        mocked_handle.assert_called_once()
-        assert result == NO_DOCUMENTS_ADDED
-
-    def test_does_not_skip_due_to_its_own_task_row(
-        self,
-        mocker: pytest_mock.MockerFixture,
-    ) -> None:
-        """
-        GIVEN:
-            - An enabled mail account with a rule
-            - A running mail fetch task row belonging to this very task
-        WHEN:
-            - Mail accounts are processed under that task id
-        THEN:
-            - The task does not skip itself and handles the account
-        """
-        PaperlessTaskFactory.create(
-            task_id="self-task-id",
-            task_type=PaperlessTask.TaskType.MAIL_FETCH,
-            trigger_source=PaperlessTask.TriggerSource.SCHEDULED,
-            status=PaperlessTask.Status.STARTED,
-        )
-
-        mocked_handle = mocker.patch.object(
-            tasks.MailAccountHandler,
-            "handle_mail_account",
-            return_value=0,
-        )
-
-        result = tasks.process_mail_accounts.apply(task_id="self-task-id").result
-
-        mocked_handle.assert_called_once()
-        assert result == NO_DOCUMENTS_ADDED