Repository navigation
Commit 44149b3
authored
Written-by: Dev
Fixes #2733
Owner: Chief
Reviewer: Reviewer
Tip SHA: d4b68fc
Base: 6bd98a7
Docs-not-needed: this changes a step's failure condition inside
`.github/actions/flybot-tests/action.yml`; no `docs/subsystems/` page
owns it (the pages' owned paths start in `core/`), and no architecture
moves.
## What changed
The `Unit tests with coverage` step ran the core suite and then only
pasted coverage into the step summary, so a glob that matched nothing
stayed green: node exits 0 with `ℹ tests 0` / `ℹ skipped 0`, the
coverage table reads `all files | 100.00 | 100.00 | 100.00`, and
`lcov.info` holds `TN:` alone while `if-no-files-found: ignore` uploads
it as an artifact.
After the `tee` and before the summary block, the step now reads the
count back out of the captured output:
```sh
tests=$(sed -n 's/^ℹ tests \([0-9]*\)$/\1/p' test-output.txt | tail -1)
echo "tests: ${tests:-?}"
[ "${tests:-0}" -ge 1 ] || { echo "::error::core suite ran no tests (glob matched nothing)"; exit 1; }
```
- A missing `ℹ tests` line falls back to `0` and fails, so truncating or
renaming the reporter output cannot buy a pass.
- `set -o pipefail` is kept and the summary's two lines still come from
the same `grep -E 'file +\| line %|all files' | sed 's/^ℹ //' | head
-2`, so what the summary shows does not move.
- The failing step prints `tests: 0` before the `::error::` annotation;
a real run prints `tests: <n>` with `n ≥ 1`.
One file changes (`+4`); `.github/workflows/test.yml` and `policy/` are
untouched, so this pull request carries no `Policy-sign:`.
### Call sites, checked before landing
`grep -rn "flybot-tests" .github/workflows` returns exactly one:
```
.github/workflows/test.yml:69: - uses: ./.github/actions/flybot-tests
```
That call passes `node: '24'`, `dsh: 0.2.1-rc.2`, `coverage: 'true'`,
`plugins: 'false'` — a real run of the 300+ file core suite, not an
intentional empty one. No call site wants an empty run, so none is
exempted and no condition was added to skip the guard.
## Verification at the tip
Tip SHA: `d4b68fc022a63bb454b201af04858e800fc90be7`
Ran: the step's `run:` block, extracted verbatim from `action.yml` at
the tip and at the base blob, executed in a scratch `core/` with
`GITHUB_STEP_SUMMARY` set.
**Red first — base blob (no guard), empty glob.** The step exits 0 with
nothing run:
```text
$ node --test … "test/*.test.js" | tee test-output.txt # test/ holds only harness-resolve.mjs
ℹ tests 0
ℹ pass 0
ℹ fail 0
ℹ skipped 0
ℹ all files | 100.00 | 100.00 | 100.00 |
BASE EXIT=0
```
**Red first — tip blob, same empty glob.** Same output up to the guard,
then it refuses:
```text
tests: 0
::error::core suite ran no tests (glob matched nothing)
TIP EXIT=1
```
**Green — tip blob, one real test file.** One matched test, exit 0, and
the summary block is byte-identical in shape to the old one:
```text
ℹ tests 1
tests: 1
TIP GREEN EXIT=0
```
and `summary.md` afterwards (same two lines the old step wrote):
### Coverage: core (ds-bot) (Node 24, DSH x)
```
file | line % | branch % | funcs % | uncovered lines
all files | 100.00 | 100.00 | 100.00 |
```
Not run (and why): `npm --prefix core test` / the full 300+ file suite
was not run on this box — `core/node_modules` is absent from a tarball
checkout and `dev/harness` has no `@deepseek-ai/*` install here, so a
local attempt died on `Cannot find module 'jsdom'` /
`ERR_MODULE_NOT_FOUND: @deepseek-ai/dsh-llm` before any test of this
change could run; that is the environment, not the diff. The full suite
is the `test` job's job: `tests-ok` at this tip is the evidence, and the
guard itself only reads that job's own output.
## Risk
T3 by shape (`.github/actions/**`), and the behavioral change is the
point: an empty core run now fails instead of reporting 100% coverage.
`merge-gate` recomputes the tier; no `risk:*` label is written by hand.
## Checkpoint
<!-- flybot:checkpoint -->
```json
{
"pr": 3053,
"claim": "Chief",
"issue": 2733,
"summary": "The core coverage step captured node's output but never read the test count, so an empty glob passed with `tests 0`, 100% coverage in the summary and a 4-byte lcov artifact. It now parses the `ℹ tests` line out of test-output.txt after the tee and fails with `::error::core suite ran no tests (glob matched nothing)` when the count is missing or below 1. The summary's two lines and the pipefail behaviour are unchanged, and the repository's only flybot-tests call site is a real run, so nothing is exempted.",
"files": [".github/actions/flybot-tests/action.yml"],
"checks": { "ran": "bash step-tip.sh / step-base.sh over an empty and a one-test glob in a scratch core/", "result": "green", "at": "d4b68fc022a63bb454b201af04858e800fc90be7" },
"next": "Wait for tests-ok on the tip; the step's own red/green is proven locally, the full suite is CI's."
}
```
<!-- /flybot:checkpoint -->
1 parent 6bd98a7 commit 44149b3
1 file changed
Lines changed: 4 additions & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
132 | 132 | | |
133 | 133 | | |
134 | 134 | | |
| 135 | + | |
| 136 | + | |
| 137 | + | |
| 138 | + | |
135 | 139 | | |
136 | 140 | | |
137 | 141 | | |
| |||
0 commit comments