paramsSignature() serialized each parameter into a joined "name:type, name:type" string, and both diffMembers() and diffSurfaces() re-split that string on ", " to detect newly-added required parameters. Any parameter whose own type text contains a comma (a two-arg generic like Map<string, number>, an inline object-type literal with multiple properties, a tuple, etc.) got re-split into fake extra "parameters", corrupting the count. Confirmed as a real false positive via a synthetic repro: adding a harmless optional b?: Map<string, number> property to an existing object-typed function parameter was misread as "gained a new required parameter" and fired the tool's own "VERIFIED BREAKING CHANGE" banner (exit 2) for a change that was 100% additive -- exactly the kind of lie this tool exists to catch, now coming from the tool itself. Fix: track each parameter as structured data (new ParamSurface: name/optional/type) throughout, never a flattened string. collectParams() replaces the string-returning paramsSignature() as the source of truth for comparisons; paramsSignature()/ formatParams() now only produce the display string, never fed back into a diff. newlyAddedRequiredParams() compares the real param lists directly. Verified: synthetic repro goes from exit 2 (false positive) to exit 0 (correct); negative control with a genuinely new required param after a comma-containing type still correctly fires exit 2; both previously-documented real-world regression checks re-run unchanged -- usesend SDK v1.6.0..v1.7.2 (73 commits, 5 structural changes, exit 0) and resend-node v4.6.0->v4.7.0 (23 structural changes, exit 2, identical before/after this change via git stash comparison). npm run build clean.
6 lines
45 B
Plaintext
6 lines
45 B
Plaintext
node_modules/
|
|
dist/
|
|
*.log
|
|
.DS_Store
|
|
.gstack/
|