[milescoviello]

What

An executable converted with --trace-imports counts its calls to the libraries in APS5_TRACE_CALLS, but only a crash or an abort report prints them, and then only the 32 most called imports. A run that ends normally, or is killed at a timeout, leaves no record of what it called.

With APS5_TRACE_CALLS_FILE set to a path, the first traced import starts a thread that rewrites that file every 10 seconds. It holds one line per import called so far, most called first: the number of calls, the library and the import name. A run that is killed keeps its last rewrite.

245557 libc.prx gQX+4GDQjpM#guest
3866 libSceAgc.prx 57labkp+rSQ

I used it to list what the 10 retail games on my test host actually call (200 to 476 different imports in 2 to 5 minutes each) and to check which of them reach a function that still has a NotImplemented_nid_no_patch path. The crash report couldn't answer that, because none of those runs crashed in a library.

How

• CallTrace.cpp: WriteCallCounts takes every import record with a nonzero count, sorts the records by count (stable, so equal counts keep their order), and writes the lines through a std::ofstream on a std::filesystem::path.
• StartCallCountWriter runs once, when the first traced import sets up the trace. It reads APS5_TRACE_CALLS_FILE, through _wgetenv on Windows as the other path variables do, and starts a detached thread that rewrites the file every 10 seconds.
• WriteCallCounts_nid_no_patch(const char* utf8Path) writes the same file on demand. It returns false for a null or empty path, or when the file can't be written. The test uses it.
• ENVIRONMENT.md, TRACING.md and the tracing section of USAGE.md describe the variable.

Tested

• Rebased onto main 17f5335, with no conflicts, as head ba73390965. guest_call_trace and guest_call_trace_abort pass 3 of 3 on Linux (Ubuntu 26.04.1, GCC 15.2, Release), and 3 of 3 on the Windows 11 VM (WinLibs GCC 15.2.0, Release) on this head and on its merge base. Conventions are clean, and the overlap scan shows no line-level overlap. The Windows runs cover the direct path-writing call; the 10-second writer thread is still exercised only by the Linux title runs below. The other results are from pre-rebase heads.

Linux (Ubuntu 26.04.1, GCC 15.2, Release), on main 0c6a5bd, then rebased onto a6dcea9. That merge doesn't touch these files.

• guest_call_trace, guest_call_trace_abort and guest_call_trace_abort_race pass 3 of 3.

• The new checks in guest_call_trace run after the existing calls and compare the written file exactly: 80001 Integers, 3 Blocks, 2 Doubles, 1 Mixed, 1 Variadic, in that order, each with the library name it was traced under. They also check that a path under a regular file, a null path and an empty path return false.

• On main the test doesn't build, because WriteCallCounts_nid_no_patch is new.

• Titles: Steins;Gate, Ghost of Tsushima, Forspoken, MGS4, Astro Bot, Crimson Desert, MXGP 26, Super Meat Boy 3D, CoD Black Ops II and Borderlands 4, converted with --trace-imports, APS5_TRACE_CALLS=all and APS5_TRACE_CALLS_FILE=calls.txt, for 120 to 300 s each. Every run left its file (204 to 476 lines), including the two that were killed (Ghost of Tsushima at 54 s, Astro Bot at 75 s). These runs used an earlier local version of this change that wrote the same lines unsorted.

• Windows 11 VM, WinLibs GCC 15.2.0, Release, head f1a1c05 and its merge base a6dcea9: guest_call_trace and guest_call_trace_abort (the two that run on Windows) pass 3 of 3 on both. So the new checks, the exact file contents included, pass there.

Not tested: macOS, and the 10-second writer thread on Windows (the test calls WriteCallCounts_nid_no_patch directly; the titles above ran on Linux).

• Used by @klosixgaming-arch (comment) on 15 titles with main 1a745ae and this branch, APS5_TRACE_CALLS=all, about 2 minutes each. It showed which census stubs and absent imports titles actually call, not just link: sceSystemServiceParamGetInt in all 13 titles that boot far enough, and timespec_get 19,588 times in one title (feat(libc): implement C11 timespec_get for PPSA34015 #9388's case).

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none
• GPUs tested: none needed (libc only)
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 9408	docs/dev/TRACING.md L14

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 1896	docs/dev/ENVIRONMENT.md
PR 3786	docs/user/USAGE.md
PR 5021	docs/user/USAGE.md
PR 5925	docs/user/USAGE.md
PR 6171	docs/user/USAGE.md
PR 6924	docs/dev/ENVIRONMENT.md docs/user/USAGE.md
PR 7449	docs/dev/ENVIRONMENT.md
PR 7457	docs/dev/ENVIRONMENT.md
PR 7514	docs/dev/ENVIRONMENT.md
PR 7604	docs/dev/ENVIRONMENT.md docs/dev/TRACING.md
PR 7642	docs/dev/ENVIRONMENT.md docs/user/USAGE.md
PR 8004	docs/dev/ENVIRONMENT.md
PR 8077	docs/dev/ENVIRONMENT.md
PR 8246	docs/dev/ENVIRONMENT.md
PR 8484	docs/dev/ENVIRONMENT.md
PR 8896	docs/dev/ENVIRONMENT.md
PR 9248	docs/dev/ENVIRONMENT.md
PR 9345	docs/dev/ENVIRONMENT.md docs/user/USAGE.md
PR 9356	docs/dev/ENVIRONMENT.md
PR 9385	docs/dev/ENVIRONMENT.md
PR 9387	docs/dev/ENVIRONMENT.md
PR 9400	docs/dev/ENVIRONMENT.md
PR 9669	docs/dev/ENVIRONMENT.md
PR 9676	docs/dev/ENVIRONMENT.md
PR 9898	docs/dev/ENVIRONMENT.md
PR 10193	docs/dev/ENVIRONMENT.md docs/user/USAGE.md
PR 10442	docs/dev/ENVIRONMENT.md
PR 10547	docs/dev/ENVIRONMENT.md

[MarcZX10]

The fixed-capacity import records remain stable while counts are sampled atomically, the output sorts every nonzero record with deterministic tie ordering, and the Linux report covers both the on-demand API and periodic file snapshots during real title runs (including runs later killed). Windows validates the direct path-writing API, while the periodic thread itself is explicitly untested there.

The branch was rebased onto 6ef5a1c4c9, but current upstream main has since advanced to 17f5335562. Please update to current main and rerun the focused guest_call_trace tests plus conventions and overlap. Since the file-path/environment handling has a Windows-specific _wgetenv branch, please also run the updated Windows focused test or state why that environment is unavailable. Hosted product jobs remain skipped. I did not run local tests.

[milescoviello]

@MarcZX10 rebased onto main 17f5335 as ba73390965, with no conflicts. guest_call_trace and guest_call_trace_abort pass 3 of 3 on Linux, and 3 of 3 on the Windows 11 VM on this head and on its merge base. Conventions are clean, and the overlap scan has no line-level entry.

On Windows the tests cover WriteCallCounts_nid_no_patch, the direct path-writing call that the writer thread repeats. The thread itself only starts when a traced title sets up its imports, and writes after 10 s. No test runs that long, and my traced title runs are on Linux, so the Windows thread remains untested, as the Tested section says.

AI-assisted: yes (Claude Code)

[klosixgaming-arch]

Used this branch (main 1a745ae38e + #9439, APS5_TRACE_CALLS=all, about 2 min per title, 15 titles) to see which census stubs and absent imports titles actually call, rather than only link.

Stubs actually called (call count in parentheses):

• sceSystemServiceParamGetInt: all 13 titles that boot far enough (1 to 8 calls each)
• kevent (nid_postfix): fc27 (1)
• _Thrd_id (nid_postfix): ds (2)
• sceAudioOut2MasteringInit: shsm (1)

Absent imports actually called:

• timespec_get: fc27 (19588 calls; the feat(libc): implement C11 timespec_get for PPSA34015 #9388 case)
• _ZSt15system_categoryv: ds (8)
• Il2Cpp entries (il2cpp_api_lookup_symbol etc.) appear as absent but are provided by the title's own module, so they are not real gaps.

Not covered: runs were not played past title/boot, so anything only reached in gameplay is missing. Titles that abort before tracing (ufc5 at sceConvertKeycodeGetImeKeyboardType, ds without APS5_PLAYGO_UNLISTED_CHUNKS) produced no or short traces. Raw per-title files available.

AI-assisted: yes. Written with AI assistance and reviewed by me.

[milescoviello]

@klosixgaming-arch thanks, that is exactly the use this was for: what titles call rather than what they link. I added it to Tested.

AI-assisted: yes (Claude Code)

[milescoviello]

@MarcZX10 rebased onto current main cf528e7 as you asked; head a938dd8f20, one commit, no conflicts.

• Linux (Ubuntu 26.04.1, GCC 15.2, Release): the 4 call_trace tests (guest_call_trace, guest_call_trace_abort, guest_call_trace_abort_race, guest_call_trace_fixed_base_abort) pass 3 of 3.
• Windows 11 VM (WinLibs GCC 15.2.0 r7, Release): the call-trace tests registered there, guest_call_trace and guest_call_trace_abort, pass 3 of 3 on the merge base and on this head. That build takes the _wgetenv branch for APS5_TRACE_CALLS_FILE, and guest_call_trace writes through the path API. As you noted, the periodic writer thread is still only exercised on Linux.
• tools/check_conventions.py passes. Overlap (tools/pr_overlap.py on this head): no open PR touches CallTrace.cpp, CallTrace.hpp or the test. perf(agc): pipeline indirect draws behind the packet worker (APS5_PIPELINE_INDIRECT) #1896, feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021 and perf(agc): build the pipelines of a submission ahead on a compile pool #5925 share docs/dev/ENVIRONMENT.md or docs/user/USAGE.md, but they conflict with main already, so the tool couldn't compare them. This PR adds its own rows and lines there.

AI-assisted: yes (Claude Code)

Dismissing a stale review: the requested changes were addressed after it (new head) and the merge queue blocks on it. Please request changes again if anything is still open.

[adriadam10]

Checked the writer against CopyCallCounts_nid_no_patch (same fixed-capacity records, acquire loads, atomic counts), the exact-file checks in guest_call_trace (they cannot build without WriteCallCounts_nid_no_patch), the null/empty/unwritable path cases and the three docs entries. The rebase onto current main that was requested is in; Conventions, relinker and write-watch are green, and the full Build runs in the merge queue; merged.

[milescoviello]

Thanks for checking the writer and the exact-file tests, and for merging, @adriadam10.

AI-assisted: yes (Claude Code)