andgineer / dinary-server / 24677965019

77%

Build Type

push

github

Committed by andgineer

Commit Message

Rebuild logging_mapping from latest_year, deduped by canonical key

Sourcing logging_mapping from year=0 — the cross-year aggregation of
every historical (sheet_category, sheet_group) pair — flooded the table
with stale capitalisations and cross-category aliases. The projection's
"first row by id" tie-breaker then routed `POST /api/expenses` writes
to whichever historical row happened to have the lowest id: e.g.
category `гаджеты` was written as `Машина / Gadgets`, category `еда`
as `Еда / ` (capitalised), category `машина` as `Машина / Car wash`.

Replace `_rebuild_logging_mapping_from_year_zero` with
`_rebuild_logging_mapping_from_latest_year`, which:

- Copies rows only from `import_mapping WHERE year = latest_year`
  (the user's current sheet conventions, kept small and predictable
  by step 8 of seed_classification_catalog).
- Dedups by canonical key `(category_id, event_id, sorted(tag_ids))`,
  so multiple legacy (sheet_category, sheet_group) pairs that resolve
  to the same 3D key collapse to a single output row.
- Synthesises the canonical default `(name, '', event=NULL, tags=[])`
  for any active category whose latest_year rows happen to all carry
  an event or tag set, so the bare API path never falls through to
  the "first row by id" branch.
- Falls back to canonical-defaults-only when no positive latest_year
  exists (fresh install with no configured import_sources).

Update architecture.md "Logging projection rules" with the new rebuild
invariant. Replace `test_logging_mapping_bootstrapped_from_year_zero`
with `_from_latest_year`, which asserts the dedup invariant and the
canonical-default safety net for every active category.

Made-with: Cursor

Coverage Stats

24 of 32 new or added lines in 1 file covered. (75.0%)

1 existing line in 1 file now uncovered.

1780 of 2482 relevant lines covered (71.72%)

0.72 hits per line