b4486492
retro: what the loud-misconfiguration pair cost
a73x 2026-08-23 10:39
Commit message
RETRO.md
| Old | New | ||
|---|---|---|---|
| @@ -29,3 +29,13 @@ One line per push to `main`: what slowed the work down. Enforced by | |||
| 29 | a tree that no longer existed. Proving the rest was comments-only needed a | 29 | a tree that no longer existed. Proving the rest was comments-only needed a |
| 30 | purpose-built token-stream comparator; go/printer reflows, so a diff that | 30 | purpose-built token-stream comparator; go/printer reflows, so a diff that |
| 31 | round-trips through it reads as changed when nothing was. | 31 | round-trips through it reads as changed when nothing was. |
| 32 | |||
| 33 | ## 2026-08-23 | ||
| 34 | |||
| 35 | - Loud misconfiguration: the survey of both call sites was done in a worktree | ||
| 36 | still parked on the previous main, so the two tests the work was told not to | ||
| 37 | duplicate looked like they did not exist and were nearly rewritten. A | ||
| 38 | worktree's branch is not the branch you are about to fork from, and nothing | ||
| 39 | says so. `make shape-check` also diffs the generated graph against the | ||
| 40 | COMMIT, so adding one import fails the gate until the regenerated docs are | ||
| 41 | committed — the same trap proto-check sets, in a second place. | ||