Adds a command-level test exercising the real zstd-unavailable branch
in document_exporter, makes document_importer only append the
"zstd archives require Python 3.14+" hint when zstd is actually among
the unreadable codecs, and adds the lzma counterpart to the
stored-level-rejection test for symmetry.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The raw library bounds (-131072, 22) come from an internal zstd
implementation constant (-ZSTD_TARGETLENGTH_MAX), not a meaningful
distinct level — deeper negative values than -22 buy nothing over -22
in practice, and the zstd CLI/community convention only uses -22..22.
Exposing the raw range via --zip-compression-level would let a user
pass a number like -50000 that "validates" but means nothing.
test_zip_lzma_compression_round_trips / test_default_zip_uses_deflate
built a real zip and read compress_type back off it — that only
reconfirms zipfile.ZipFile() honors its own compression= kwarg
(ZipExportSink's own tests already cover that forwarding). What this
command owns is resolving the --zip-compression string to the right
zipfile constant; assert that resolution directly against the mocked
ZipExportSink construction call instead. Drops the sample-doc copytree
setup those tests needed only to build a real archive.
test_compression_method_is_applied_to_file_entries built a real zip and
read back compress_type — but that only confirms zipfile.ZipFile()
honors its own compression= kwarg, which is documented stdlib behavior,
not something our code could get wrong. ZipExportSink's only
responsibility is forwarding compression/compresslevel unchanged to
ZipFile(); assert that directly against the mocked constructor call.
test_compressing_method_beats_stored asserted that DEFLATED produces a
smaller archive than STORED — that's zlib's job, not ours.
test_compression_method_is_applied_to_file_entries already covers what
our code is actually responsible for: threading the requested
compression method through to the zip entry's compress_type.
The design is now implemented (src/documents/export/sinks.py); the
planning docs served their purpose. The deferred zip-compression-control
follow-up plan/spec remains, since that work hasn't landed yet.
Path.rename() raises FileExistsError on Windows when the destination
already exists; Path.replace() is atomic cross-platform. Also widen
document_exporter's sink parameters from the concrete
DirectoryExportSink | ZipExportSink union to the ExportSink ABC, so a
future sink implementation is a pure addition rather than requiring
every call site's annotation to change.
DirectoryExportSink.add_json and _commit_streamed_file each inlined the
same hash-and-compare logic; extracted _content_unchanged(). Replaced
ZipExportSink._ensure_dirs's repeated slice/join with a prefix
accumulator and hoisted the _zip-is-open assertion out of the loop. No
behavioral change.
The abstract method had no return annotation, so pyrefly inferred -> None
and flagged both DirectoryExportSink.stream and ZipExportSink.stream as
incompatible overrides.
The implementation plan for Task 2 diverged from the design spec
(export-sink-architecture-design.md), leaving ExportSink as a plain
class with NotImplementedError bodies instead of the specified
AbstractContextManager subclass. Use abc.ABC + @abstractmethod so a
concrete sink missing a required method fails at instantiation
rather than at first call.
This branch is scoped to the export sink architecture + zip
compression control work; drop the search, bulk-edit,
ingestion-staging, vector-store, and other unrelated design docs
and scratch profiling scripts that had accumulated on it.
Adds the extensibility design that hoists the bulk-edit operation
definition (today smeared across serialisers.py, views.py, bulk_edit.py
and keyed three different ways) behind a BulkEditOperation registry +
PermissionRequirements value object, with per-operation OpenAPI examples.
Contract-preserving refactor; both docs reviewed across multiple passes
(permission matrix verified, both view call sites accounted for).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the search-error-shapes stub with a full design spec and a TDD
implementation plan for friendlier advanced-search error messages.
Empirically validated against a live Tantivy index: three error families
(UnknownFieldError, InvalidFieldValueError, MalformedQueryError),
proactive numeric validation plus a parse_query backstop, comparison
operators confirmed working, and a parse-based field drift guard.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
stumpylogandClaude Sonnet 5
