perf: reduce apply() bytecode size (Phase 1 + 2: TempLocalCountVisitor, buffer 256→32, TAILCALL trampoline extraction)#650

Merged

fglock merged 13 commits intomasterfrom

perf/reduce-apply-bytecode

May 2, 2026

Commits on May 2, 2026

perf: reduce apply() bytecode by improving TempLocalCountVisitor accuracy

fglock
and
devin-ai-integration[bot]
committed
docs: add evaluation guide and Phase 1 progress tracking to reduce-apply-bytecode

fglock
and
devin-ai-integration[bot]
committed
perf: Phase 2 - extract per-call-site TAILCALL trampoline to resolveTailCalls()

fglock
and
devin-ai-integration[bot]
committed
perf: fix Phase 2 hot-path regression — move resolveTailCalls() inside control-flow branch

fglock
and
devin-ai-integration[bot]
committed
fix: eliminate VerifyError from int-typed callContextSlot in JVM bytecode emitter

fglock
and
devin-ai-integration[bot]
committed
docs: update design doc with Phase 2 bugfix for callContextSlot VerifyError

fglock
and
devin-ai-integration[bot]
committed
docs: record DBIx::Class full-suite PASS as safe revert point

fglock
and
devin-ai-integration[bot]
committed
fix: restore correct tied-scalar FETCH count for fallback compound assigns

fglock
and
devin-ai-integration[bot]
committed
perf: Phase 4 — route oversized eval-generated methods to interpreter

fglock
and
devin-ai-integration[bot]
committed
refactor: clean up Phase 4 comment/structure in EmitterMethodCreator

fglock
and
devin-ai-integration[bot]
committed
docs: expand Phase 4 gate comment so developers don't remove it

fglock
and
devin-ai-integration[bot]
committed
docs: replace box header in Phase 4 comment with plain text

fglock
and
devin-ai-integration[bot]
committed
fix: overload fallback — use overload; (no ops) should not throw

fglock
and
devin-ai-integration[bot]
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf: reduce apply() bytecode size (Phase 1 + 2: TempLocalCountVisitor, buffer 256→32, TAILCALL trampoline extraction)#650

perf: reduce apply() bytecode size (Phase 1 + 2: TempLocalCountVisitor, buffer 256→32, TAILCALL trampoline extraction)#650
fglock merged 13 commits intomasterfrom
perf/reduce-apply-bytecode

Commits on May 2, 2026

perf: reduce apply() bytecode by improving TempLocalCountVisitor accuracy

docs: add evaluation guide and Phase 1 progress tracking to reduce-apply-bytecode

perf: Phase 2 - extract per-call-site TAILCALL trampoline to resolveTailCalls()

perf: fix Phase 2 hot-path regression — move resolveTailCalls() inside control-flow branch

fix: eliminate VerifyError from int-typed callContextSlot in JVM bytecode emitter

docs: update design doc with Phase 2 bugfix for callContextSlot VerifyError

docs: record DBIx::Class full-suite PASS as safe revert point

fix: restore correct tied-scalar FETCH count for fallback compound assigns

perf: Phase 4 — route oversized eval-generated methods to interpreter

refactor: clean up Phase 4 comment/structure in EmitterMethodCreator

docs: expand Phase 4 gate comment so developers don't remove it

docs: replace box header in Phase 4 comment with plain text

fix: overload fallback — use overload; (no ops) should not throw