test(headless): read both ways an adapter names a mounted repo file - #2316
Merged
Conversation
The check that holds the mount declarations to what the adapters actually read is worth exactly what its reader sees, and the competitor half of it saw less than it looked. It scanned for bare filenames and resolved them against `packages/headless/harbor`, so a read of any repo file outside that one directory matched nothing and passed — `dist/index.js`, which the Maka adapter already reads, would have been invisible had a competitor adapter reached for it. This is the gap that cost `run-host-cell.mjs`: the Maka side had no authority check at all, `install()` probed for a file nobody declared, and a missing probe target aborts the trial before the arm runs. Closing it for Maka left the same weakness on the other side. Both halves now share one reader that matches the two forms adapters actually write — the mounted path spelled out in full, and the same path joined onto the mount root — and it is exercised directly rather than only through whichever forms today's adapters happen to use.
…nothing
Two gaps the first round left, both of the same kind: a regex guard that
stops matching still passes.
The reader recognised the join form only when it hung off `maka_repo`,
and `maka_agent.py` hangs its own headless CLI path off a
`Path("/opt/maka-agent")` constant instead — that read was invisible, and
covered only by the directory mount it happens to fall inside. Match the
chain rather than what it hangs off: the root expression varies per
adapter and nothing stops the next one from binding its own name, while
a repo path necessarily starts at a repo directory.
And nothing held the reader to seeing anything at all. Rewriting the
adapters into a form it does not parse left the read set empty and the
suite green — the check silently no longer checking, which is exactly how
its predecessor rotted. An adapter with files declared for it is one
whose source names them, so an empty read set there is now a failure.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Each arm is mounted exactly the repo files it declares, and the check that holds those declarations to what the adapters actually read is the only thing standing between a missed entry and a trial that aborts inside the container. #2298 added that check for the Maka adapter after
run-host-cell.mjs— probed byinstall()in every cell-mode branch — was left out of the declaration. It left the competitor half as it was.The competitor half saw less than it looked like it saw. It scanned adapter sources for bare filenames and resolved them against
packages/headless/harbor, so it could only ever notice a read of a file in that one directory. A competitor adapter reaching forpackages/headless/dist/index.js— a path the Maka adapter already reads — matched nothing and passed. It was also matching by coincidence in the other direction: any string literal equal to some filename underharbor/counted as a read of it.Both halves now share one reader that matches the two forms adapters actually write:
codex_agent.py's_DEEPSEEK_MODELS_PATH),opencode_agent.py,maka_agent.py).Bare-filename matching is gone; both forms above cover every read in the adapters today, so nothing is lost with it. The reader is also exercised directly, so it is no longer worth only what today's adapters happen to spell.
Refs #2298.
Verification
opencode_agent.pyatpackages/headless/dist/index.js(a path outsideharbor/, undeclared, unmounted). Onmainall 19 mount tests pass — the gap, reproduced. With this change:opencode_agent.py names packages/headless/dist/index.js, which opencode is not mounted.@maka/headless1431 pass / 0 fail.npm run lintandnpm run formatclean.Not run: desktop E2E and Storybook — test-only change in headless.
Review focus
The reader is a regex over Python source, which is a heuristic, not a parser: an adapter that builds a container path through a variable or an f-string is still invisible to it. That limit is unchanged from #2298; what changes here is that the limit is now the same on both sides and is stated in one place instead of being an accident of which directory the matcher scanned.