moonbitlang / core / 6711

89%

push

github

perf(builtin): size Map::filter's result for the entries kept

Map::filter (#4253) allocated the result with the source's capacity to
avoid growing. That was never needed for hash reuse: entries store the
full hash, so it fits a table of any size. Presizing only bought a
sparser table, at up to 2x the memory (100x for a 1% filter), and was
slower than exact sizing for selective filters because the whole table
had to be allocated.

Collect the kept entries, then build a table sized like Map([...])
would size it. The saving over a comprehension is now only what filter
itself provides: no rehashing and no tuple allocation.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

8 of 8 new or added lines in 1 file covered. (100.0%)

16420 of 18354 relevant lines covered (89.46%)

274442.97 hits per line