commit:?<root> token-diffs the whole initial tree (O(repo), seconds)Loading the parentless ROOT commit (commit:?c7184818) takes ~4.4s because it
token-level weave-diffs the ENTIRE initial-import tree as all-INS, where a
one-sided (vs-empty) diff should be a cheap raw all-add. Split from BLAME-006
(the perf half). Method: Issues.
be commit:?<sha> (beagle/BE.cli.c:332 + relay :453-459) runs graf
diff:?<sha>; graf_first_parent_hex returns GRAFNONE for the parentless root
(GRAF.exe.c:75), so it falls to GRAF.exe.c:505 → GRAFDiffTreeRefs
(graf/DIFFREF.c:564) which TOKEN-diffs every one of the 291 initial-import
files as all-INS (@@ -1,0 +1,N @@) — O(repo), tokenizing+weave-diffing all.commit:?<root> (and any one-sided / vs-empty diff) loads in time proportional
to the commit, not the repo — emit raw all-add without token-level weave diff
when one side is empty.DIFFREF.c:564 / the WEAVE diff core), detect a one-empty-side
diff and short-circuit to a raw all-INS emit (skip tokenize+LCS) for each file.be commit:?<root> vs a normal commit; confirm O(repo) hot path
in GRAFDiffTreeRefs.8d66369c (FF child of 3dc19b0a; msg BLAME-006b: skip token-diff of
binary blobs). Scan's "291-file all-INS" premise was wrong — the 4.35s was one
9.5 MB binary (dog/dogs.png) tokenized in 3.17s. Fix (graf): WEAVEFromBlobRm
(one-sided/vs-empty short-circuit to raw all-add) + diffref_is_binary (git's
NUL-in-first-8000 heuristic) so binaries are never token-diffed; root commit-show
4.35s→1.52s, text output byte-identical. Test: Case I in beagle/test/be-diff-projector.sh.
302/302. NB landing: reconciled a graf/DIFFREF.c overlap with DIFF-002 (both survive,
verified on trunk); the POSTCFLCT marker false-positive on graf/WEAVE.c required the
documented --force conflict-scan bypass (see POST-006/POST-017).