Interpreter startup dominates a small-input CLI run

Index: Known gaps

Status (re-verified 2026-10-09 at 456989f): OPEN. Measurement only. The entry says import cost was never investigated, and no lazy-import work is recorded; the Click registration-by-import pattern is unchanged. unverified: wall time was not re-measured.

abicheck --version — interpreter plus import plus Click tree construction, before any work at all — costs 0.56–0.66 s. A stored-snapshot/stored-snapshot compare of a small fixture costs ~1.05 s end to end, so roughly 60% of it is startup. compare --dry-run (startup plus config and input resolution, diff never run) costs 0.60–0.79 s, which bounds resolution itself at well under 0.2 s.

Consequences worth knowing before optimizing anything else: a change that halves the actual L2 comparison work would move that run's wall time by under 20%, and the full-CLI lane's absolute noise floor has to be set around this cost rather than around the comparison's (hence --regress-min-delta-seconds 0.5 by default there, where the in-process gates use 0). A warm-cache dump is not measurably faster than a cold one on a small fixture for the same reason — see below.

Not investigated: which imports dominate, and whether lazy-importing the heavy ones is feasible without breaking the registration-by-import-side-effect pattern the CLI is built on.