ci: the drift job checks an installed wheel, not just the source tree - #129
Merged
Merged
Conversation
Closes #103. #116 already covered most of what that issue asked for -- a weekly resolve that ignores `uv.lock`, reporting to the issue tracker, not gating pull requests. What it did not cover is the criterion's other half: importing every subpackage from the *built wheel*. The distinction is the one #101 was: `mcp>=1.2` resolved to 2.0.0 for anyone installing fresh, `mcp.server.fastmcp` had gone, and `grapharc[mcp]` was broken on arrival while every locked job stayed green. A suite run from the source tree does not see a packaging break that only shows in an installed wheel -- a subpackage dropped from the build, or an extra whose new major moves a module the package imports at import time. The walk moves out of `ci.yml`'s heredoc into `scripts/wheel_import_walk.py`, because two jobs now need exactly this check and two copies would drift apart. Same logic, two parameters: which checkout to compare against, and the prefix the import must come from. `scripts/` holds no `grapharc` package, so running it by path does not put the checkout on `sys.path` -- the property the old `cd /tmp` was there for, kept and now asserted with a message rather than a bare `assert`. Verified by running it for real rather than reasoning about it, and it earned its keep immediately: against the wheel then sitting in `dist/` it reported `grapharc.examples.plan_research` missing -- correctly, because that wheel was built before #115 merged. Rebuilt, it walks 132 modules clean. Deliberately **not** done: putting a ceiling on the eight unbounded extras. The issue asks for a decision on that and argues a bound should be "a ceiling with a reason attached, not a reflex" -- so adding eight of them at once is the reflex it warns against, and it would refuse users upgrades that are fine. Detection is what this closes; the policy is left open. Verified: 2214 selected, 13 deselected, ruff clean; the sdist's required-files and secret-leak checks still pass with `scripts/` added. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This was referenced Sep 26, 2026
Shashankss1205
added a commit
that referenced
this pull request
Sep 26, 2026
#115 and #129 merged after the release commit. 0.1.8 is still unpublished and untagged, so folding them into its notes is right for the same reason #124 and #128 did it. The research registry gets a full entry: it is a shipped module and it closes the gap that the planner's headline capability -- decomposition -- was the one with no runnable example, while the only registry in the box demonstrated refusal. Credited to @only-ouc, whose pull request it is. The drift job's wheel walk goes in the tooling paragraph beside the styling gate: it changes nothing shipped, it changes what a break in shipped behaviour would be caught by. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
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.
Closes #103.
#116 already delivered most of that issue's acceptance criteria: a weekly resolve that ignores
uv.lock, takes the newest release of every extra, reports to the issue tracker, and does not gate pull requests. What it did not cover is the other half — "imports every subpackage from the built wheel".Why that half matters separately
It is the shape #101 actually was:
mcp>=1.2resolved to2.0.0for anyone installing fresh,mcp.server.fastmcpno longer existed there, andgrapharc[mcp]was broken on arrival — while every job resolving from the lockfile stayed green.A suite run from the source tree does not see that. It misses a packaging break that only appears in an installed wheel: a subpackage dropped from the build, or an extra whose new major moves a module the package imports at import time.
So
upstream-driftnow builds the wheel, installs it with[all]into a clean environment resolved frompyproject.toml's ranges, and walks it.The walk is now shared, not duplicated
It moves out of
ci.yml's heredoc intoscripts/wheel_import_walk.py. Two jobs need exactly this check —buildagainst whatuv.lockpins,upstream-driftagainst what upstream has released since — and two copies would drift apart, which is the same failure mode as the phase vocabulary in #117.Same logic, two parameters: which checkout to compare against, and the path prefix the import must come from.
scripts/contains nographarcpackage, so running the file by path does not put the checkout onsys.path— the property the oldcd /tmpexisted for, preserved, and now reported with a message instead of a bareassert.It earned its keep on the first run
Run against the wheel that was sitting in
dist/, it reported:Correct — that wheel predated #115. Rebuilt, it walks 132 modules clean. (Worth noting for the pending 0.1.8 upload:
dist/needed rebuilding again after #115, and has been.)Deliberately not done
No ceilings on the eight unbounded extras. The issue asks for a decision there and argues a bound should be "a ceiling with a reason attached, not a reflex" — adding eight at once is precisely that reflex, and it would refuse users upgrades that are perfectly fine. This PR closes the detection gap, which is what the acceptance criteria specify; the pinning policy stays an open decision for you.
Verified
2,214 selected, 13 deselected,
ruff check .clean. The sdist's required-files and secret-leak checks still pass withscripts/added (it is not shipped — CI reads it from the checkout).🤖 Generated with Claude Code