Sequential --stop-on-failure writes no report files and skips the rerun cache #1415

Problem

A sequential --stop-on-failure run writes none of the requested report files and does not update .bashunit/last-failed. The same run under --parallel writes all of them.

Reproduced on main 776997e9, macOS arm64, Bash 3.2.57:

printf 'function test_first_ok() { assert_same 1 1; }\nfunction test_second_ko() { assert_same 1 2; }\nfunction test_third_ok() { assert_same 1 1; }\n' > stop_test.sh
rm -rf .bashunit
./bashunit --skip-env-file -S --log-junit junit.xml --report-json report.json stop_test.sh
ls junit.xml report.json .bashunit/last-failed
./bashunit --skip-env-file -S --parallel --log-junit junit.xml --report-json report.json stop_test.sh
ls junit.xml report.json .bashunit/last-failed

After the first run ls reports all three paths missing. After the parallel run all three exist and last-failed holds stop_test.sh:test_second_ko. Both runs print the failure, Stop on failure enabled... and exit 1.

Where

• src/runner/diagnostics.sh:69-77: sequential mode leaves through exit "$EXIT_CODE_STOP_ON_FAILURE".
• src/main/run.sh:83 routes that exit to bashunit::main::handle_stop_on_failure_sync (src/main/run.sh:405-426), which prints the summary and cleans up, but never reaches the report writers (exec_tests, src/main/run.sh:181-256), bashunit::rerun::persist (:298) or the coverage, snapshot and GitHub annotation steps around them.
• Parallel mode only marks a flag file, so exec_tests runs to its normal end and writes everything.

Why it matters

--stop-on-failure --log-junit is a common CI pairing. The step fails, and the report the CI reads is missing or stale from the previous run. --rerun-failed after a stopped run replays the previous cache instead of the test that just stopped the run; if the previous run was green, it reports "No previously failing tests recorded" and runs the whole suite.

Suggested change

Make the sequential stop path return to exec_tests instead of exiting from inside the test loop: for example set a flag in halt_if_stop_on_failure for sequential runs too, have call_test_functions and load_test_files break out when it is set, and let exec_tests render, write reports, persist the rerun cache and exit 1 as it already does for parallel runs. If the exit path has to stay, call the same report, rerun and cleanup steps from handle_stop_on_failure_sync.

Acceptance criteria

• Sequential -S with --log-junit, --report-json, --report-html, --report-tap, --report-md, --log-gha writes each requested file and it holds the tests that ran.
• Sequential -S writes the stopping test to .bashunit/last-failed, and --rerun-failed replays it.
• The console output, exit code and tear_down_after_script behaviour of -S are unchanged.
• An acceptance test compares the files written by -S in both modes and fails on the current code.

Proposed fix

Make the sequential stop leave the test loop by a flag instead of an exit, so exec_tests reaches its own report, rerun and cleanup steps:

1. bashunit::runner::halt_if_stop_on_failure (src/runner/diagnostics.sh:69-77): for sequential runs set _BASHUNIT_STOP_REQUESTED=true (declare it next to _BASHUNIT_FILE_TEARDOWN_PENDING in src/runner/hooks.sh:351) instead of exit "$EXIT_CODE_STOP_ON_FAILURE".
2. bashunit::runner::call_test_functions (src/runner/exec.sh:173-176): also break when [ "${_BASHUNIT_STOP_REQUESTED:-}" = true ], so the remaining functions of the file are skipped.
3. bashunit::runner::load_test_files (src/runner/discovery.sh:59): at the top of the loop [ "${_BASHUNIT_STOP_REQUESTED:-}" = true ] && break. The file that stopped already ran its tear_down_after_script at :267 on the way out.
4. bashunit::main::exec_tests (src/main/run.sh:165-168): print Stop on failure enabled... when the flag is set, not only under --parallel, so the console output keeps its line and position; render_result then exits 1 as today.
5. Delete the EXIT trap at src/main/run.sh:83 and bashunit::main::handle_stop_on_failure_sync (:405-426); keep run_pending_file_teardown, which Ctrl-C still needs (--stop-on-failure skips tear_down_after_script in sequential runs #1321, Ctrl-C does not run tear_down_after_script for the file in flight #1323).

Bash 3.0: plain globals and [ ] tests; nothing version specific. Public behaviour that must not change: the console output of a sequential -S run (pinned by tests/acceptance/snapshots/bashunit_stop_on_failure_test_sh.*), exit code 1, tear_down_after_script running before the summary. Docs: docs/command-line.md --stop-on-failure entry, add one sentence that reports and .bashunit/last-failed are written for the tests that ran. CHANGELOG: one line under Unreleased, Fixed.

Agent brief

• Branch: fix/1415-stop-on-failure-writes-reports from fresh origin/main. Commit: fix(runner): write reports and the rerun cache when --stop-on-failure stops a sequential run with Closes #1415 in the body.
• Files: src/runner/diagnostics.sh, src/runner/exec.sh, src/runner/discovery.sh, src/runner/hooks.sh, src/main/run.sh; extend tests/acceptance/bashunit_stop_on_failure_test.sh (copy test_bashunit_when_stop_on_failure_option).
• RED first:

function test_stop_on_failure_writes_the_junit_report_and_the_rerun_cache_sequentially() {
  local test_file=./tests/acceptance/fixtures/test_bashunit_when_stop_on_failure.sh
  local report cache_dir
  report="$(bashunit::temp_file junit).xml"
  cache_dir="$(bashunit::temp_dir rerun)"
  BASHUNIT_RERUN_CACHE_DIR="$cache_dir" ./bashunit --no-parallel --env "$TEST_ENV_FILE" \
    --stop-on-failure --log-junit "$report" "$test_file" >/dev/null 2>&1 || true
  assert_file_exists "$report"
  assert_file_exists "$cache_dir/last-failed"
}

Both assertions fail on the current code. Run only that file: ./bashunit tests/acceptance/bashunit_stop_on_failure_test.sh.

• Non-goals: the parallel stop path (flag file) already works; do not change the summary text, exit codes, or --rerun-failed semantics (A green run of a subset clears .bashunit/last-failed #1418).
• Gate, once, when the change is complete: ./bashunit tests/, ./bashunit --parallel --simple --strict tests/, make sa, make lint. make lint is the formatting authority; never shfmt -w; 2-space indent.
• Done when: every acceptance box is checked, the gate is green, CHANGELOG and docs are updated, and a PR is open with the pr skill.