CodSpeedHQ/valgrind-codspeed - CodSpeed

valgrind-codspeed

Blog Docs Changelog

Performance History

Latest Results

chore: release 3.26.0-0codspeed3 Co-Authored-By: Claude <noreply@anthropic.com>

master

9 hours ago

test(callgrind): add regression tests for runtime obj-skip - runtime_obj_skip_c: a fn from a skipped object must not leak into the dump as a top-level fn= block when it is the first BB after START (the cxt==0 force-push path) - runtime_obj_skip_underflow: a RET past an empty call stack (handleUnderflow) must not re-leak the skipped fn -- the Python 3.14 deep recursive interpreter-dispatch shape Both exercise --obj-skip / CALLGRIND_ADD_OBJ_SKIP from a separately-linked .so. Also filters callgrind diagnostic logs from test stderr and gitignores the test build artifacts (binaries, .so, logs).

master

1 day ago

test(callgrind): add regression tests for runtime obj-skip - runtime_obj_skip_c: a fn from a skipped object must not leak into the dump as a top-level fn= block when it is the first BB after START (the cxt==0 force-push path) - runtime_obj_skip_underflow: a RET past an empty call stack (handleUnderflow) must not re-leak the skipped fn -- the Python 3.14 deep recursive interpreter-dispatch shape Both exercise --obj-skip / CALLGRIND_ADD_OBJ_SKIP from a separately-linked .so. Also filters callgrind diagnostic logs from test stderr and gitignores the test build artifacts (binaries, .so, logs).

cod-2714-investigate-pytest-flamegraph-regression-final

1 day ago

fix(callgrind): skip push_cxt for innermost frame in shadow-stack seeder When CALLGRIND_START_INSTRUMENTATION fires from a constructor (e.g. via LD_PRELOAD or __attribute__((constructor))), the innermost frame (frame 0) returns before the workload runs. Seeding it with push_cxt made it the root of the cost attribution tree, holding ~100% of inclusive cost. Fix: frame 0 now gets only an SP entry (for correct SP-based unwinding) but no push_cxt. Ancestor frames (1..n-1) continue to be fully seeded. For JIT/trampoline cases where frame 0 continues executing the workload, setup_bbcc auto-pushes its context on the first BB after START, so skipping the seed is safe for both scenarios. Also removes temporary RECON debug prints from the seeder.

cod-2714-investigate-pytest-flamegraph-regression-final

2 days ago

test(callgrind): add regression tests for runtime obj-skip - runtime_obj_skip_c: a fn from a skipped object must not leak into the dump as a top-level fn= block when it is the first BB after START (the cxt==0 force-push path) - runtime_obj_skip_underflow: a RET past an empty call stack (handleUnderflow) must not re-leak the skipped fn -- the Python 3.14 deep recursive interpreter-dispatch shape Both exercise --obj-skip / CALLGRIND_ADD_OBJ_SKIP from a separately-linked .so. Also filters callgrind diagnostic logs from test stderr and gitignores the test build artifacts (binaries, .so, logs).

cod-2714-investigate-pytest-flamegraph-regression-final

2 days ago

fix(callgrind): skip push_cxt for innermost frame in shadow-stack seeder When CALLGRIND_START_INSTRUMENTATION fires from a constructor (e.g. via LD_PRELOAD or __attribute__((constructor))), the innermost frame (frame 0) returns before the workload runs. Seeding it with push_cxt made it the root of the cost attribution tree, holding ~100% of inclusive cost. Fix: frame 0 now gets only an SP entry (for correct SP-based unwinding) but no push_cxt. Ancestor frames (1..n-1) continue to be fully seeded. For JIT/trampoline cases where frame 0 continues executing the workload, setup_bbcc auto-pushes its context on the first BB after START, so skipping the seed is safe for both scenarios. Also removes temporary RECON debug prints from the seeder.

cod-2714-investigate-pytest-flamegraph-regression-final

2 days ago

2 days ago

2 days ago

Latest Branches

-73%

fix(callgrind): initialize stack on instrumentation start#15

6 days ago

ae9ee5c

cod-2714-investigate-pytest-flamegraph-regression-final

N/A

refactor(bench): migrate benchmarks to codspeed exec CLI#16

2 days ago

d2dd609

cod-2766-migrate-valgrind-benchmarks-to-exec-cli

+14%

fix: flamegraph regression#14

8 days ago

cd1d3d4

cod-2714-investigate-pytest-flamegraph-regression-3

© 2026 CodSpeed Technology

Home Terms Privacy Docs