Issues

regex: a construction still pays two address-keyed side-table inserts (2.0M/reply) and 72 MB of headers per reply

#9908

· proggeramlug opened

on Sep 6, 2026

main no longer boots Claude-of-Duty: "Set.prototype.add is not a function" in world construction

#9907

· proggeramlug opened

on Sep 6, 2026

YoungLogWalk: `table_len` and `logged` mean different quantities in closure.dynamic_props' two walks, so young_log_an.py compares unlike numbers

#9897

· proggeramlug opened

on Sep 6, 2026

[gc-restore-coverage] reports `dirty_pages` while the walk iterates `dirty_old_pages`, and counts objects where the cost is per slot

#9877

· proggeramlug opened

on Sep 6, 2026

diag: [gc-primitive-dispatch] string_wrappers counter has no writer on main since #9810/#9814 (reads 0 unconditionally)

#9874

· proggeramlug opened

on Sep 6, 2026

lint: gc_runtime_root_holders.py is red on main 35c36f425 — PASS1_MARKED's policy.rs pin went stale in the train; SHAPE_CACHE_YOUNG re-add hazard for #9838

#9873

· proggeramlug opened

on Sep 6, 2026

gc: negative-cache the unregistered 1 MiB classes — they are 90.7% of every classification miss left, and only 0.14% are rejected by the span check

#9852

· proggeramlug opened

on Sep 6, 2026

gc: adaptive tenuring picks survival age 1 and pays 2-5 unproductive full collections per claude-code turn for it

#9851

· proggeramlug opened

on Sep 6, 2026

codegen: match the Intl.Segmenter for-of that owns 60-85% of claude-code's turn, and count whether the tier fires

#9843

· proggeramlug opened

on Sep 6, 2026

GC: #9755's young-entry logs are a net 0.78x on claude-code, and closure.dynamic_props is a 2.56x regression — the logs do not drain because the scanner predicate keeps Longlived

#9841

· proggeramlug opened

on Sep 6, 2026

gc: the base ArenaBytes arm paces the nursery by a quantity that is 99 % old generation (#7909's young basis, not applied to the base arm)

#9839

· proggeramlug opened

on Sep 6, 2026

perf(gc): per-object layout metadata is address-keyed, so every copying minor rehashes it — 304 MB allocated / 14.9 MB live per 400-char reply

#9792

· proggeramlug opened

on Sep 5, 2026

Navigation Menu

Uh oh!

Uh oh!

All issues

regex: a construction still pays two address-keyed side-table inserts (2.0M/reply) and 72 MB of headers per reply

main no longer boots Claude-of-Duty: "Set.prototype.add is not a function" in world construction

YoungLogWalk: `table_len` and `logged` mean different quantities in closure.dynamic_props' two walks, so young_log_an.py compares unlike numbers

[gc-restore-coverage] reports `dirty_pages` while the walk iterates `dirty_old_pages`, and counts objects where the cost is per slot

diag: [gc-primitive-dispatch] string_wrappers counter has no writer on main since #9810/#9814 (reads 0 unconditionally)

lint: gc_runtime_root_holders.py is red on main 35c36f425 — PASS1_MARKED's policy.rs pin went stale in the train; SHAPE_CACHE_YOUNG re-add hazard for #9838

gc: negative-cache the unregistered 1 MiB classes — they are 90.7% of every classification miss left, and only 0.14% are rejected by the span check

gc: adaptive tenuring picks survival age 1 and pays 2-5 unproductive full collections per claude-code turn for it

codegen: match the Intl.Segmenter for-of that owns 60-85% of claude-code's turn, and count whether the tier fires

GC: #9755's young-entry logs are a net 0.78x on claude-code, and closure.dynamic_props is a 2.56x regression — the logs do not drain because the scanner predicate keeps Longlived

gc: the base ArenaBytes arm paces the nursery by a quantity that is 99 % old generation (#7909's young basis, not applied to the base arm)

perf(gc): per-object layout metadata is address-keyed, so every copying minor rehashes it — 304 MB allocated / 14.9 MB live per 400-char reply

Uh oh!

Uh oh!

Search results