You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 1975c27
Browse filesBrowse the repository at this point in the historyBrowse files
test(requesty): pin timeout normalisation, timeout-covered lookup and the three cancel paths
Three review asks on this head, all in the spec so no production behaviour moved.
createOptions timeout: an it.each now asserts what the SDK is actually handed for two
values that whole numbers cannot distinguish - 1500.5 must be ceiled to 1501 and 3e9 must be
clamped to 2147483647. Measured: with toSdkTimeoutMs replaced by a pass-through of
options.timeoutMs both cases fail (2 failed) and the received value is 1500.5, which is the
sentence that turns the coverage ask into a proof: dropping the conversion is observable.
Model lookup under timeoutMs: a new case runs completePrompt with { timeoutMs: 20 } and no
caller signal at all against a getModels that never settles, so the only cancellation source
is the timeout this method configures; it asserts the provider AbortError and that create was
never reached. The control that drives the lookup race from bridgedCallerSignal instead was
attempted twice and the vitest process was killed both times, so that half is reported as
unmeasured rather than claimed.
Detach lifecycle: the resolve test and the failure test are folded into one it.each with a
third case that aborts mid-flight. The fold asserts nothing new about detaching for the two
original outcomes; the new ground is that dispose also runs on the cancellation exit, which
neither original pinned.
The comment that said it mirrored the Requesty test inside the Requesty spec no longer claims
a sibling: git grep for withSettleGuard/notifyCreateStarted across src/api/providers/__tests__
returns matches only in this file (28), so there was no other provider test to name and the
comparison was dropped rather than re-pointed.
Verified: tsc -p tsconfig.json --noEmit clean; spec 50 passed (was 45 - the fold nets one
extra case and the timeout it.each adds two, the lookup case one); prettier --check clean;
eslint over the touched file clean after formatting; pre-commit hook run by hand before the
commit. src/eslint-suppressions.json untouched, no changeset, no CHANGELOG.
0 commit comments