spp test / test_single_module.sh: a failing run prints no TEST RESULTS block and ./spp t exits 0

[gonzalesedwin1123]

Found while running the negative control for #383 / PR #525 (reverting the fix, expecting red): ./spp t spp_base_common exited 0 and printed nothing after "Running tests (this may take a few minutes)...", while the log said 1 failed, 1 error(s) of 10 tests. The CLAUDE.md pitfall "./spp t can exit 0 even when the run failed; read the log" describes the symptom; this is the mechanism.

Mechanism (three separate defects, scripts/test_single_module.sh + spp)

1. set -e aborts the script before the results are parsed. The script runs under set -e (line 21). run_tests_docker() correctly wraps docker compose run in set +e / set -e, captures TEST_EXIT_CODE and returns it (lines 193-217; run_tests_local() does the same). But the call site is a bare statement:

   if [ "$MODE" = "docker" ]; then
       run_tests_docker        # line 514: non-zero return under set -e => script exits here
   else
       run_tests_local
   fi
   parse_and_display_results   # line 520: never reached when odoo-bin exited non-zero
   exit $?

   odoo-bin exits non-zero whenever a test fails or the registry fails to load, so exactly the runs that matter skip parse_and_display_results: no TEST RESULTS header, no result line, no --- Test failures --- excerpt, no CRITICAL: line. A green run prints the block; a red run prints nothing. (The test DB cleanup inside the function still runs because it precedes the return.)

2. parse_and_display_results never returns non-zero. Its last statement is echo "", so exit $? at line 521 is always exit 0, even when it just printed Failed: 2 or CRITICAL: ... Failed to load registry. The odoo-bin exit code captured in step 1 is never propagated.

3. spp discards the script's status. spp (cmd_test, ~line 363) calls run(cmd, check=False) and does not sys.exit with the returned code, so ./spp t is exit 0 regardless of what the script did.

Net effect: CI is unaffected (the workflow greps the log's N failed, M error(s) line itself), but every local/agent run has to open /tmp/openspp-test-logs/<module>_unittest_*.log and grep odoo.tests.result to learn the verdict, and any tooling that trusts the exit code (subagents, scripts, && chains) sees success on failure.

Proposed fix

• Call the runner under set +e (or if ! run_tests_docker; then ...; fi, or run_tests_docker || TEST_EXIT_CODE=$?) so the script always reaches parse_and_display_results.
• Make parse_and_display_results return 1 when FAILED > 0, TEST_ERRORS > 0, CRITICAL_ERROR is set, or no result line was found at all (the "No tests were executed" branch), and return 0 otherwise; exit with that.
• In spp, sys.exit(run(cmd, check=False).returncode) so ./spp t mirrors the script.
• Optional: when the result line is missing but the log exists, print the last 20 log lines so the failure is visible without opening the file.

Repro

Any module with a failing test: e.g. on a branch where a test asserts something false, run ./spp t <module>; observe exit 0 and no TEST RESULTS block; grep odoo.tests.result /tmp/openspp-test-logs/<module>_unittest_*.log | tail -1 shows the failures. Logs from the #525 negative control: /tmp/openspp-test-logs/spp_base_common_unittest_20260918_103209.log (local machine).