feat(valgrind): declare the pid a benchmark ran in

[lvaroqui]

Make the valgrind instrument use the pid argument of set_executed_benchmark, which it ignored until now.

When the pid differs from the calling process, the instrument registers a Benchmark pid: <pid> desc line just before the URI dump. It does this through the new CALLGRIND_ADD_DESC request from CodSpeedHQ/valgrind-codspeed#43, so the line lands in that benchmark's part header:

part: 2
desc: Spawned pid: 289126
desc: Benchmark pid: 289126
desc: Trigger: Client Request: exec_harness::true_twice

This lets a profile consumer tell a process that only launched the benchmark apart from the one that ran it. A launcher's own spawn-and-wait cost can then be left out.

The public API is unchanged. Benchmarks run in the calling process, which covers every integration today, get no extra line and no extra part. On a valgrind without the request, the call only logs a warning and the dump happens as before.

Notes for review:

• The desc is formatted into a stack buffer, so an allocation failure can't skip the URI dump.
• dist/core.c is regenerated with Zig 0.14.1, the version that reproduces the current file byte-for-byte; the diff is mostly renumbered transpiler symbols.
• check-callgrind-header fails until feat(callgrind): add CALLGRIND_ADD_DESC client request valgrind-codspeed#43 merges.

Closes COD-3722

[greptile-apps]

Confidence Score: 4/5

[Medium risk] Adds process ID tracking to benchmark profiling output.

The PR appears safe to merge, with a non-blocking gap in regression coverage for the new profile-header behavior.

Findings

1.  PID header lacks regression coverage ▶

Fix with agent prompt

### Issue 1
src/instruments/valgrind.zig:50-53
The new branch adds `Benchmark pid` to the Callgrind part header, but the existing Valgrind smoke test does not inspect a Callgrind profile, and the no-crash test checks only the return code. A missing or misplaced description could therefore pass CI, leaving profile consumers unable to distinguish launcher cost from benchmark cost. A test that inspects the dumped header would protect this behavior.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

The PR adds a Callgrind client request that labels a dumped part with the benchmark PID when it differs from the reporting process, and regenerates the distributed C implementation.

• Updates the Valgrind header, C wrapper, and Zig instrument while preserving the existing public API.
• Switches the lint workflow to a pinned prek action.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A["set_executed_benchmark(pid, URI)"] --> B{"pid differs from getpid()?"}
  B -- Yes --> C["Register Benchmark pid description"]
  B -- No --> D["Dump stats at URI"]
  C --> D
  D --> E["Callgrind part header and counters"]

Loading

_{Reviews (1) · Last reviewed commit: "ci: use prek instead of pre-commit (#32)"}