Fix: handle single-class classifier in threshold prediction (#14471) (#14473)

scikit-learn treats a single-class fit as binary, so predict_proba returns
[1 - p, p] for one class and argmax could index past classes_. Use only
column 0, the real class, so the confidence threshold still applies.

src/documents/classifier.py

@@ -54,6 +54,10 @@ def _predict_with_threshold(classifier, X, threshold: float) -> int | None:
     cutoff so that uncertain predictions are discarded rather than assigned.
     """
     probas = classifier.predict_proba(X)[0]
+    if len(classifier.classes_) == 1:
+        # scikit-learn treats a single-class fit as binary, so predict_proba
+        # returns [1 - p, p] for one class. Column 0 is the real class.
+        probas = probas[:1]
     best_idx = int(probas.argmax())
     best_class = int(classifier.classes_[best_idx])
 

src/documents/management/commands/document_exporter.py

@@ -3,6 +3,7 @@ from itertools import islice
 from pathlib import Path
 from typing import TYPE_CHECKING
 from typing import Any
+from typing import Final
 
 from allauth.mfa.models import Authenticator
 from allauth.socialaccount.models import SocialAccount
@@ -70,6 +71,15 @@ from paperless.models import ApplicationConfiguration
 from paperless_mail.models import MailAccount
 from paperless_mail.models import MailRule
 
+# Derived names replace the original's extension with these suffixes
+THUMBNAIL_SUFFIX: Final[str] = "-thumbnail.webp"
+ARCHIVE_SUFFIX: Final[str] = "-archive.pdf"
+# Leave room for the longest derived suffix, which replaces the original's extension
+MAX_BASE_NAME_BYTES: Final[int] = Document.MAX_PUBLIC_FILENAME_BYTES - max(
+    len(THUMBNAIL_SUFFIX),
+    len(ARCHIVE_SUFFIX),
+)
+
 
 def serialize_queryset_batched(
     queryset: "QuerySet[Any]",
@@ -512,7 +522,10 @@ class Command(CryptMixin, PaperlessCommand):
                     counter=filename_counter,
                 )
             else:
-                base_name = document.get_public_filename(counter=filename_counter)
+                base_name = document.get_public_filename(
+                    counter=filename_counter,
+                    max_bytes=MAX_BASE_NAME_BYTES,
+                )
 
             # The thumbnail and archive names are derived from the stem, so
             # uniqueness must hold without the original's extension
@@ -541,7 +554,7 @@ class Command(CryptMixin, PaperlessCommand):
         document_dict[EXPORTER_FILE_NAME] = original_arc
 
         if not self.no_thumbnail:
-            thumbnail_name = base_name.parent / (base_name.stem + "-thumbnail.webp")
+            thumbnail_name = base_name.parent / (base_name.stem + THUMBNAIL_SUFFIX)
             if self.use_folder_prefix:
                 thumbnail_name = Path("thumbnails") / thumbnail_name
             thumbnail_arc = thumbnail_name.as_posix()
@@ -550,7 +563,7 @@ class Command(CryptMixin, PaperlessCommand):
             thumbnail_arc = None
 
         if not self.no_archive and document.has_archive_version:
-            archive_name = base_name.parent / (base_name.stem + "-archive.pdf")
+            archive_name = base_name.parent / (base_name.stem + ARCHIVE_SUFFIX)
             if self.use_folder_prefix:
                 archive_name = Path("archive") / archive_name
             archive_arc = archive_name.as_posix()

src/documents/models.py

@@ -157,6 +157,8 @@ class StoragePath(MatchingModel):
 
 class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-missing]
     MAX_STORED_FILENAME_LENGTH: Final[int] = 1024
+    # Per-component limit of common filesystems (ext4, XFS, btrfs, ZFS, APFS)
+    MAX_PUBLIC_FILENAME_BYTES: Final[int] = 255
 
     correspondent = models.ForeignKey(
         Correspondent,
@@ -484,28 +486,37 @@ class Document(SoftDeleteModel, ModelWithOwner):  # type: ignore[django-manager-
     def archive_file(self):
         return Path(self.archive_path).open("rb")
 
-    def get_public_filename(self, *, archive=False, counter=0, suffix=None) -> str:
+    def get_public_filename(
+        self,
+        *,
+        archive: bool = False,
+        counter: int = 0,
+        max_bytes: int = MAX_PUBLIC_FILENAME_BYTES,
+    ) -> str:
         """
         Returns a sanitized filename for the document, not including any paths.
+
+        The name fits in max_bytes (UTF-8). A long title is shortened before
+        the counter and extension are appended, because sanitize_filename
+        truncates from the end. The tail is cut only when max_bytes is
+        smaller than the tail itself.
         """
         # Root owns metadata for all versions
         context_document = (
             self.root_document if self.root_document_id is not None else self
         )
-        result = str(context_document)
 
-        if counter:
+        extension = ".pdf" if archive else self.file_type
-            result += f"_{counter:02}"
+        tail = f"_{counter:02}{extension}" if counter else extension
 
-        if suffix:
+        head_bytes = max(0, max_bytes - len(tail.encode()))
-            result += suffix
+        head = str(context_document).encode()[:head_bytes].decode(errors="ignore")
 
-        if archive:
+        return pathvalidate.sanitize_filename(
-            result += ".pdf"
+            head + tail,
-        else:
+            replacement_text="-",
-            result += self.file_type
+            max_len=max_bytes,
-
+        )
-        return pathvalidate.sanitize_filename(result, replacement_text="-")
 
     @property
     def file_type(self):

src/documents/tests/test_classifier.py

@@ -14,6 +14,7 @@ from django.test import override_settings
 from django.test.utils import CaptureQueriesContext
 from pytest_django.fixtures import Settings
 from pytest_mock import MockerFixture
+from sklearn.neural_network import MLPClassifier
 
 from documents.classifier import ClassifierModelCorruptError
 from documents.classifier import DocumentClassifier
@@ -1036,6 +1037,49 @@ def test_classifier_match_threshold_default() -> None:
     assert settings.CLASSIFIER_MATCH_THRESHOLD == 0.3
 
 
+class TestPredictWithThresholdSingleClass:
+    @pytest.fixture
+    def single_class_model(self) -> tuple[MLPClassifier, np.ndarray]:
+        X = np.random.default_rng(0).random((2, 5))
+        return MLPClassifier(tol=0.01, random_state=0).fit(X, [3, 3]), X[:1]
+
+    def test_returns_only_class(
+        self,
+        single_class_model: tuple[MLPClassifier, np.ndarray],
+    ) -> None:
+        """
+        GIVEN:
+            - A classifier fitted on a single class (every document shares one label)
+        WHEN:
+            - A prediction is requested with the threshold disabled
+        THEN:
+            - No error is raised and the only class is returned
+        """
+        classifier, X = single_class_model
+
+        assert _predict_with_threshold(classifier, X, threshold=0.0) == 3
+
+    def test_applies_threshold(
+        self,
+        single_class_model: tuple[MLPClassifier, np.ndarray],
+    ) -> None:
+        """
+        GIVEN:
+            - A classifier fitted on a single class
+        WHEN:
+            - A prediction is requested
+        THEN:
+            - The class's confidence is still compared against the threshold
+        """
+        classifier, X = single_class_model
+        confidence = classifier.predict_proba(X)[0][0]
+
+        assert _predict_with_threshold(classifier, X, threshold=confidence - 0.01) == 3
+        assert (
+            _predict_with_threshold(classifier, X, threshold=confidence + 0.01) is None
+        )
+
+
 class TestPreprocessContent:
     @pytest.fixture
     def samples(self) -> Path:

src/documents/tests/test_document_model.py

@@ -4,13 +4,16 @@ from datetime import date
 from pathlib import Path
 from unittest import mock
 
+import pytest
 from django.test import TestCase
 from django.test import override_settings
 from faker import Faker
+from pytest_mock import MockerFixture
 
 from documents.models import Correspondent
 from documents.models import Document
 from documents.tasks import empty_trash
+from paperless_testing.factories import DocumentFactory
 
 
 class TestDocument(TestCase):
@@ -257,3 +260,72 @@ def test_suggestion_content() -> None:
     )
     doc.content = content_over_limit
     assert doc.suggestion_content == expected_cropped_content
+
+
+class TestPublicFilenameLength:
+    @pytest.fixture(autouse=True)
+    def _no_parser_registry(self, mocker: MockerFixture) -> None:
+        # The parser registry reads ApplicationConfiguration from the database
+        extensions = {"application/pdf": ".pdf", "image/jpeg": ".jpg"}
+        mocker.patch(
+            "documents.models.get_default_file_extension",
+            side_effect=extensions.__getitem__,
+        )
+
+    LIMIT = Document.MAX_PUBLIC_FILENAME_BYTES
+
+    @pytest.mark.parametrize(
+        ("title", "archive", "counter", "max_bytes", "tail"),
+        [
+            pytest.param("a" * 300, False, 0, LIMIT, ".jpg", id="ascii"),
+            pytest.param("a" * 300, True, 0, LIMIT, ".pdf", id="archive"),
+            pytest.param("ä" * 200, False, 0, LIMIT, ".jpg", id="umlaut"),
+            pytest.param("ä" * 200, False, 1, LIMIT, "_01.jpg", id="count"),
+            pytest.param("a" * 300, False, 1, 100, "_01.jpg", id="max"),
+        ],
+    )
+    def test_long_title_keeps_tail_within_limit(
+        self,
+        title: str,
+        *,
+        archive: bool,
+        counter: int,
+        max_bytes: int,
+        tail: str,
+    ) -> None:
+        """
+        GIVEN:
+            - A document whose title alone is longer than the byte limit
+        WHEN:
+            - The public file name is generated
+        THEN:
+            - The name ends with the counter and extension
+            - The name fits in the limit
+        """
+        name = DocumentFactory.build(
+            title=title,
+            mime_type="image/jpeg",
+        ).get_public_filename(
+            archive=archive,
+            counter=counter,
+            max_bytes=max_bytes,
+        )
+
+        assert name.endswith(tail)
+        assert len(name.encode()) <= max_bytes
+
+    def test_long_title_counters_give_distinct_names(self) -> None:
+        """
+        GIVEN:
+            - A document whose title alone is longer than the byte limit
+        WHEN:
+            - Public file names are generated with increasing counters
+        THEN:
+            - Each counter produces a distinct name
+        """
+        doc = DocumentFactory.build(title="ä" * 200, mime_type="application/pdf")
+
+        names = [doc.get_public_filename(counter=counter) for counter in range(3)]
+
+        assert len(set(names)) == 3
+        assert all(len(n.encode()) <= self.LIMIT for n in names)

src/documents/tests/test_management_exporter.py

@@ -1537,3 +1537,70 @@ class TestExportDerivedFileNameCollisions:
             assert (
                 target / record[EXPORTER_ARCHIVE_NAME]
             ).read_text() == f"archive {pk}"
+
+    def test_long_titles_export_within_file_name_limit(
+        self,
+        paperless_dirs: PaperlessDirs,
+        tmp_path: Path,
+    ) -> None:
+        """
+        GIVEN:
+            - Two documents with the same created date and the same title
+            - The title alone is longer than 255 bytes and contains umlauts
+        WHEN:
+            - The exporter runs
+        THEN:
+            - The export finishes with distinct original, thumbnail and archive names
+            - Every exported file name fits in the byte limit
+            - Each exported file holds its own document's content
+        """
+        for pk in (1, 2):
+            (paperless_dirs.originals_dir / f"{pk:07}.pdf").write_bytes(
+                f"original {pk}".encode(),
+            )
+            (paperless_dirs.archive_dir / f"{pk:07}.pdf").write_bytes(
+                f"archive {pk}".encode(),
+            )
+            (paperless_dirs.thumbnail_dir / f"{pk:07}.webp").write_bytes(
+                f"thumbnail {pk}".encode(),
+            )
+            DocumentFactory(
+                pk=pk,
+                title="Bestellbestätigung " * 15,
+                created=date(2026, 10, 15),
+                filename=f"{pk:07}.pdf",
+                archive_filename=f"{pk:07}.pdf",
+                mime_type="application/pdf",
+            )
+
+        target = tmp_path / "export"
+        target.mkdir()
+        call_command(
+            "document_exporter",
+            str(target),
+            "--no-progress-bar",
+            skip_checks=True,
+        )
+
+        manifest = json.loads((target / "manifest.json").read_text())
+        records = {r["pk"]: r for r in manifest if r["model"] == "documents.document"}
+
+        for key in (
+            EXPORTER_FILE_NAME,
+            EXPORTER_THUMBNAIL_NAME,
+            EXPORTER_ARCHIVE_NAME,
+        ):
+            assert records[1][key] != records[2][key]
+
+        for pk in (1, 2):
+            record = records[pk]
+            for key, content in (
+                (EXPORTER_FILE_NAME, f"original {pk}"),
+                (EXPORTER_THUMBNAIL_NAME, f"thumbnail {pk}"),
+                (EXPORTER_ARCHIVE_NAME, f"archive {pk}"),
+            ):
+                assert (
+                    len(Path(record[key]).name.encode())
+                    <= Document.MAX_PUBLIC_FILENAME_BYTES
+                )
+                assert (target / record[key]).read_text() == content