[FIX] Give each upload its own temp file so concurrent tests stop clobbering each other's results

[cfsmp3]

Problem

upload_type_request() saved every upload to TempFiles/<uploaded name>:

temp_path = os.path.join(temp_dir, filename)   # filename comes from the upload
uploaded_file.save(temp_path)
# ... hash the file ...
os.rename(temp_path, final_path)

That name is derived from the regression test output, so it is identical for every test running the same regression test. Two tests running at once share one path:

1. A saves TempFiles/X.srt
2. B overwrites TempFiles/X.srt
3. A hashes it and renames it to TestResults/<hashA>.srt
4. B's rename → FileNotFoundError → HTTP 500

The 500 means no TestResultFile row is written. mod_test/controllers.py then sets got = 'error' for any regression test with no result files but expected output, which renders as "No output generated but there should be". The contributor sees their PR breaking tests it never touched.

There is a quieter variant: if B's save lands between A's save and A's hashing, A records the hash of B's output and its verdict silently flips. No error is raised in that case at all.

Evidence

From ccextractor PR #2309. Tests 9402 (master) and 9410 (the PR) ran concurrently on Linux, 15:32–16:02:

run	HTTP 500s in VM log	spurious "No output"	genuine baseline
9402 master Linux	37	36	24
9410 PR Linux	21	21	24
9403 master Windows	0	0	24
9411 PR Windows	0	0	24

The two Windows runs of the same commits were staggered (15:35 vs 15:44), hit zero 500s, and produced identical verdicts. The two Linux runs disagreed on 57 tests, in two completely disjoint sets — the signature of a race, not a code difference.

Server-side traceback:

File "mod_ci/controllers.py", line 2680, in upload_type_request
    os.rename(temp_path, final_path)
FileNotFoundError: [Errno 2] No such file or directory:
  '/repository/TempFiles/3c77a844….srt' -> '/repository/TestResults/ca8d0dd4….srt'

error.log contains 2184 of these failed renames, so this has been corrupting results for a long time. It only became obvious once a backlog started launching two tests at the same moment.

Fix

Allocate a unique temp file per upload with tempfile.mkstemp(), and finish with os.replace() (atomic). A finally block removes the temp file if anything fails, so failures no longer leak into TempFiles.

Testing

• test_upload_type_request_uses_unique_temp_path — two uploads of the same filename must get different temp paths, both results land, TempFiles ends empty. Fails on master with '…/TempFiles/shared_name.srt' == '…/TempFiles/shared_name.srt'.
• test_upload_type_request_cleans_temp_file_on_failure — a failing save leaves nothing behind.
• Updated test_upload_type_request for the new call shape (mkstemp + os.replace).

tests.test_ci.test_controllers: 202 tests, OK. isort, pydocstyle clean; mypy's only complaint is the pre-existing missing PyYAML stub, which CI installs via --install-types.

Note

This does not change how many tests genuinely fail — the baseline of 24 real failures ("Path is empty", CCExtractor exit code 10) is untouched. It stops the platform inventing extra ones and blaming them on whichever PR happened to be running.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud