[amar-python]

What

A T&E re-upload that fails every row now changes nothing: the earlier load's registry record survives, and any prior table of the same name is kept. Closes #215.

The bug

upload_te deleted the earlier load's registry record, and dropped a same-named dynamic table, before inserting any rows. It also skipped the #170 all-failed rollback for re-uploads (and not is_reupload). So a re-upload where every row failed replaced "6 loaded" with "0 loaded, 10 failed" while the 6 rows stayed in the table, and returned status: ok.

#170's original scenario still reproduced on main (same root cause, so fixed here). Before this change, on main: load a dynamic file, then upload an all-invalid T&E file with the same filename:

	before (main at ad39ddf)	after
T&E response	ok, inserted 0	error, inserted 0
prior dynamic table	dropped (404)	readable (200)
registry	replaced by te / 0 rows	still dynamic / 1 row

The fix

In api/services/te_loader.py (net -3 lines):

• The delete-permission check for overwriting stays up front, so a role without delete rights still cannot insert through a re-upload.
• _cleanup_prior_registry_entries now runs only after at least one row has landed.
• The all-failed rollback applies to re-uploads too.

Behaviour change (approved on #215)

A re-upload where every row fails now returns status: error (the same response a first-time all-failed upload already gets), not ok with 0 inserted. Re-uploads that insert at least one row behave as before.

Tests

• New tests/test_te_reupload_registry.py (3 tests, live PostgreSQL): all-failed re-upload keeps the registry record and the table agrees with it; a re-upload that inserts rows still replaces the record (guards against over-correcting); RV-F3 (BR-34): all-failed T&E import runs registry cleanup and removes prior dataset #170's prior-dynamic-table case. Two of the three failed on main for the stated reason before the fix; the third (replace-on-insert) is a guard that passes both ways.
• Two existing tests changed, deliberately: test_te_reupload_same_file_replaces_registry (test_api_coverage.py) and test_reupload_cleans_prior_registry_entry (test_te_loader_integration.py). Both re-uploaded an identical one-row file whose unique name made every row fail, so they were asserting the old behaviour by accident. They now re-upload a file that can succeed, which keeps their actual purpose (a re-upload replaces the record rather than adding a second).

Verification

Full suite against PostgreSQL 16 with all four environments deployed, as CI does: 391 tests, 388 passed, 0 failed, 0 skipped, 3 errors. The 3 errors are test_csv_loader_arbitrary_shapes (build/config.local.env not present); they fail identically on main without this change. Not run: CI (this is a draft).

Heads-up

#214 moves the test files into category folders, including the two I edited. Whichever merges second will need a small conflict resolution in those files, and my new test file will need moving into tests/system_integration/.

---

Generated by Claude Code

[amar-python]

CI: free-tier failed at Test catalogue check (docs/TEST_CASES.md is stale), not at any test. This PR adds tests, and the catalogue is generated and checked for freshness. Reproduced locally with python3 scripts/test_catalogue.py --check, fixed by regenerating it in b1593c4, check now exits 0. The lint step and the unit/regression suite in the same job had already passed.

While there I gave the three new tests a one-line docstring and made the edited test's docstring open with a complete sentence, since the catalogue shows only the first line. Waiting on the re-run.

---

Claude Code

---

Generated by Claude Code