1. Most time: waiting on the gate, since an emit_*.nika change runs the whole tests/compiler package at both settings. After that, the commit: commit.nika refused with git add failed because I had already run git rm on b4’s attachments, and git add can’t take a path that is deleted from both the index and the tree. Unstaging the deletions by name fixed it, at the cost of one turn.
2. Looked up: where the typed let writes its type. That is emit_stmt_annotation (emit_stmt.nika:435), not :260 as the brief said. I also had to find how reference and spelled combine in the parameter loop: in the named form reference is "" and spelled already holds &'r Vec<..>. One line in the brief, “the named form arrives as spelled = "&'r Vec<..>" with reference == ""”, would have saved the reading at :597-660.
3. Checks: a probe on the old binary (useful, and it gave me the test’s failure before the fix), one build cycle plus one fixpoint rebuild, the probe on the new binary, a grep of the it tests for the old lowering, and the gate. None were wasted. A shell typo that left the markdown fence in the probe cost one call.
4. Biggest saving: builder.md should say that commit.nika takes a deletion only when it is unstaged: delete with rm, not git rm, and list the path under Files:. The brief’s git rm instruction contradicts the script as written.

This site is open source. Improve this page.