Split from #14322 (item 4). The empty migrated database is already made once a run, not once a process (test/helpers/preload.ts:25, src/kernel/db.ts). What is left is to measure, not assume: in the files that boot a forge or spawn processes, how much of each test's time is setup (git repo init, seeded repos, app checkouts) the helpers redo per test.
Do: time the shared helpers' setup across the forge-booting test files (the verification's per-file times are now kept, src/checks/times-store.ts); for any setup over ~100 ms that is the same every time, build it once per run keyed by what it is built from (as the migrated database is keyed by the migration hash) and copy it per test.
Done when: the before/after seconds of the affected files are in the pull, and a test proves a changed key (a new migration, a changed seed) builds a fresh template instead of reusing the stale one.
Split from #14322 (item 4). The empty migrated database is already made once a run, not once a process (
test/helpers/preload.ts:25,src/kernel/db.ts). What is left is to measure, not assume: in the files that boot a forge or spawn processes, how much of each test's time is setup (git repo init, seeded repos, app checkouts) the helpers redo per test.Do: time the shared helpers' setup across the forge-booting test files (the verification's per-file times are now kept,
src/checks/times-store.ts); for any setup over ~100 ms that is the same every time, build it once per run keyed by what it is built from (as the migrated database is keyed by the migration hash) and copy it per test.Done when: the before/after seconds of the affected files are in the pull, and a test proves a changed key (a new migration, a changed seed) builds a fresh template instead of reusing the stale one.