Skip to content

Promote develop to main for v0.4.1 - #52

Merged
its-janghoon merged 2 commits into
mainfrom
develop
Sep 21, 2026
Merged

its-janghoon merged 2 commits into
mainfrom
develop

Conversation

@its-janghoon

Copy link
Copy Markdown
Contributor

Promotes the tree v0.4.1 was cut from. All three changes are test-only — no shipped behaviour differs from v0.4.0.

  • #49 — the slow-platform allowance now reaches the call sites that write their own bound. v0.4.0 was cut before this landed, which is why it is here.
  • #50 — the ACP subprocess waits are scaled too. Third place a fixed millisecond figure met a slower platform.
  • #51XDG_DATA_HOME is isolated for core tests, so three ModelsDev tests stop depending on whether the developer has logged in. They were green in CI and red on any machine with a real auth.json.

Release v0.4.1 is published: 21 assets, Windows installer signed with the org Authenticode certificate, macOS notarized.

its-janghoon and others added 2 commits September 21, 2026 15:31
… into core tests

Three ModelsDev tests were green in CI and red on any developer machine where
somebody had run `redrob providers login`. They clear `REDROB_API_KEY` and then
assert the static fallback catalog is served -- but `resolveApiKey()` reads THREE
sources in order: that env var, the Integration store, and `auth.json` under
`Global.Path.data`. The suite stubs the second and clears the first. The third was
the developer's own credential.

So the service found a key, fetched, and returned the live listing's figures --
`limit.output: 64000` where the fallback publishes `32000`. The assertion was
reporting the difference between two catalogues rather than the thing it claims to
check, and nothing in the test named the variable it depended on.

`XDG_DATA_HOME` is pointed at a fresh temp directory in the preload, which is
early enough: `Global.Path.data` is computed once at module load, so redirecting
it afterwards is impossible -- the suite's own `REDROB_TEST_HOME` seam is a getter
and covers `home` only, not the data directory where credentials live.

Fixed in the test setup rather than by adding a seam to production code. Nothing
about the CLI's real path resolution is wrong, and `XDG_DATA_HOME` is the standard
way to say where that directory is.

Verified by removing only the isolation: exactly those three fail again, and
nothing else in the 974 does.
…a-dir

Isolate the XDG data directory so a real login cannot leak into core tests
@its-janghoon
its-janghoon merged commit 6918fe4 into main Sep 21, 2026
38 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant