index/perma.js:122:Ab,145,157 historyOf/blobHistory/blobsFor and
index/hook.js:258:bn do a readCommit plus a tree descent per commit. On
~/src/linux kernel/sched/core.c: blobHistory 8240 ms vs 40 ms for the
identical 1940-blob scope off dag.pathRevs' REV-BLOB span; mintRef calls
it once per ref, unmemoised, INSIDE git commit. hashlet60 carries the 60
bits mintHashlet needs, so the index is a complete substitute. Found in the
2026-09-18 index review.
Serve blob history off the REV-BLOB span; memoize per (path, tip) within a mint run.
hashlet60 IS an object name), not in full sha1s: the
REV-BLOB val carries 60 bits and a 10-char hashlet reaches
exactly those, so shaBits reads either spelling and
nothing has to be named in full to be weighed. The clash
test compares the same 15, so a 60-bit twin now mints an
ambiguous hashlet where it used to mint none — no hashlet
of any length tells such a pair apart anyway.blobsFor rides the same span: it sieves the revs and only
a MATCH costs a readCommit, so the follow drops from one
tree descent per commit to one or two ODB reads.bee http cannot answer off a history that has moved.aY.historyOf stays as the ODB leg's commit list and
is the fallback's only reader now.