Remove the remaining duplicate implementations: annotation, snpdb, cross-app #1801

[davmlaw]

🤖 Written by Claude

Addresses #1801, the last of the three (after #1798 and #1800). Net -142 lines, no migration (makemigrations --check clean across all apps).

_ExternalDump (annotation/external_annotation.py)

dump_external_annotation_runs and dump_existing_annotation_runs differ only in how they get hold of their AnnotationRuns — one creates them from range locks, the other adopts existing CREATED ones. Everything after that was duplicated: the too-small rejection and revert, the dump_count == 0 skip, the lazily-computed version identity, the sidecar write, and the whole write_snakemake_bundle / re-dispatch / logging tail.

Both now build a _ExternalDump, feed each run to add_run(), and return finish(). The two log lines that named the flavour ("Dumped external ..." / "Dumped existing ...") keep their wording via the described_as argument.

FamilyGroupMixin (snpdb/models/models_cohort.py)

Ten identical methods across Trio and Quad — permissions, preview, genome_build, data_archived, get_samples, and the member-details properties. __str__ was the same bar the printed class name, so it now uses type(self).__name__ and gives the same "Trio 5" / "Quad 5". mother_details / father_details / sibling_details were three copies of one two-line shape, now _member_details(member, affected).

The fields deliberately stay where they are — related_name differs per model (trio_mother vs quad_mother), so they can't move to a shared abstract model without a migration. This is a plain mixin, so there is no schema change at all.

FamilyGroupListColumns / NamedCollectionColumns (snpdb/grids.py)

QuadsListColumns was TriosListColumns plus two sibling columns; both now declare their model, their UserGridConfig name, and their family members. CustomColumnsCollectionColumns and TagColorsCollectionColumns were identical bar the model, and are now two-line subclasses.

Verified the generated column lists are byte-for-byte what the old classes produced, in the same order.

plot_chromosome_bin_values (library/graphs/chromosomes_graph.py)

The chromosome density and homozygosity percent graphs each built the same pcolormesh band by hand — the dimension juggling, the y_top/y_bottom padding, the masked array, set_clim. It moves next to plot_chromosomes, which both already use; the caller passes its own vmax (computed vs. a flat 100.0), which was the only real difference.

as_table / as_p (snpdb/templatetags/model_helpers.py)

The shared row loop becomes _display_rows. Each filter keeps its own return type — as_p returns mark_safe, as_table a plain str because its callers apply |safe at the template — and I checked both still return SafeString / str respectively against a real model.

annotation/fake_data.py

_variant_ids_by_gene and _zipf_weight were byte-identical in analysis/fake_variant_tags.py and classification/fake_reclassifications.py. Both are create_fake_data subcommands; the shared copy sits in annotation (which owns the models the query needs) so neither app has to import the other. Moved out of #1798 at review request.

popoverHoverStay (global.js)

The hover-stay handler with its 300ms .popover:hover grace check was bound in two places. Edited in static_files/default_static, not the collectstatic output.

Identified, deliberately not changed

• annotation/tasks/calculate_sample_stats.py repeats the same ~15-line counter-bump three times. It is real duplication, but it is the innermost per-variant loop of a task that walks the whole database, and a helper call per bucket per variant is a throughput cost I didn't think worth paying for tidiness. Written up in the issue so it is on record.
• ClassificationAttachment.get_file_dict / PatientAttachment.get_file_dict — identical bar URL names, but sharing them properly means an abstract attachment model spanning two apps plus migrations. Bigger than the rest of this, and left for its own change.
• Test fixtures — annotation/tests/ has ~6 copies of the same setUpTestData. Worth tidying, but sharing fixtures is a different exercise to deduplicating production code.

Testing

• manage.py test --keepdb (whole suite) — 2426 tests, OK
• manage.py check clean; makemigrations --check reports no changes for any app
• ruff shows no new findings on any changed file
• Spot-checked the refactors against live objects: Trio/Quad mixin resolution and preview_if_url_visible, the four grids' column lists, and the two template filters' return types
• global.js has no node/eslint available here, so I checked the edit by inspection — the brace and paren deltas against master are symmetric (-4/-4, -13/-13)

🤖 Generated with Claude Code

https://claude.ai/code/session_01TKZV4Gh7ZXRjqx7ibPDTSd