codemap 0.0.22 — pack measured against git history, and walking both ways
R1-C65 — pack said nothing about what it did with the seed. An unknown name, a
.md path or an attribute id returned the global pack byte for byte, exit 0, under
a header naming the seed. Now every pack carries seeds, unreachable_from_seeds
and coverage, and a seed set that resolves to nothing is a refusal (exit 1).
R1-C66 — a seeded pack walks both ways by default. Against git history (seed the
pack with a changed file on the parent's graph, count the co-changed files in its
first ten) the one-way pack found 18.9 % on 108 commits of bquant, below grep by
names, and 0 % of co-changed tests. Under rules committed before the run, the
two-way walk found 41.6 % there and 59.9 % on 83 held-out commits of codemap.
`--direction uses` keeps the old walk. The agent A/B stays closed: against grep
the interval includes zero on one repository.
Also: R1-C59 (fewer edges on the same input is named by diff and build -o),
R1-C60 (dead-code declares nested definitions it cannot judge), R1-C64
(codemap doctor: does each dependency answer a known question correctly).
Schema 0.13 (unchanged, eleventh release). ANSWER_FORMAT 1 -> 2, only pack
changed shape; every other answer changes by its trailer alone.
Verified from PyPI against 0.0.21 on a tree carrying each defect, PYTHONHASHSEED=7:
pack --seed seed 0.0.21: caller rank 0.0 (filler), no walk named
0.0.22: caller rank 0.117, tests reached, walk: both
pack --seed no_such 0.0.21: exit 0, global pack -> 0.0.22: exit 1, refused
doctor 0.0.21: no such command -> 0.0.22: exit 0, both OK
diff, 7 of 38 calls cut 0.0.21: silent -> 0.0.22: calls 38 -> 31 named
graph nodes and edges hash identical, 0.13 -> 0.13
bquant live tree, check one line: answer format 1 -> 2; exit 0 -> 0