[dantdmdarealguy]

What

Four rows of docs/dev/ENVIRONMENT.md do not match the code on main:

• APS5_CRASH_INDIRECT_MEMORY: the row says the variable dumps the words the faulting registers point at. The Windows crash report always prints those [reg] lines (core/libs/prx/libc/src/specifics/windows/CrashReport.cpp L398-402). A non-empty value adds, for each pointer in those lines, up to 64 bytes read from where it points, as [[reg+off]] @addr lines (L403-413). That is what USAGE.md and GAME_TESTING.md already say. An empty value leaves it off (GetEnvironmentVariableA(..., nullptr, 0) > 1).
• APS5_LOG_TIMESTAMPS: the row says "milliseconds since the first log". LogMacros.hpp L19-29 prints [+S.UUUUUU s] , seconds with six microsecond digits. The start time is a static inside an inline header function, so on Windows each library (DLL) has its own copy and its own clock, started at that library's first stamped line.
• ANYPS5_LANGUAGE: the Source column names libSceSystemService/Export.cpp. The variable is read in SystemService.hpp L41 (SystemServiceConsoleLanguage, which Export.cpp and libSceNpTrophy2 call). tools/env_reference.py --table also gives SystemService.hpp.
• APS5_LABEL_FLUSH_US: the Default column says 1000. DeferredLabels.cpp L62 uses 1000 only when batched label submission is on and 250 when APS5_NO_LABEL_BATCH_SUBMIT or APS5_NO_LABEL_BATCHING turns it off (LabelBatchSubmit(), L26-29). Because the default depends on that, the Default cell is left blank and the Description states both values, as the APS5_HANDOFF_SPIN_US row does.

Tested

Windows 11 Pro, Intel Core i9-14900K, 128 GB RAM, NVIDIA GeForce RTX 5080 (driver 32.0.16.1742), WinLibs GCC 15.2.0, Release build of main 138182635 with BUILD_TESTING=ON.

• APS5_CRASH_INDIRECT_MEMORY, with tests/windows_backtrace_report_tests.exe fault and build/core/libs/libs/unpatched on PATH:
  • Unset: 2 [reg] word lines ([rdx], [rsi]), 0 [[reg+off]] lines.
  • =1: the same 2 [reg] lines plus [[rsi+0x8]] @000001de4e460150 ... and [[rsi+0x28]] @000001de4e460150 ..., 8 words (64 bytes) each.
  • Set to an empty value: 2 [reg] lines, 0 [[reg+off]] lines.
• APS5_LOG_TIMESTAMPS, with ScreenTester5 (PPSA69420) relinked by this build (relinker.exe --windows source/eboot.elf run/app.exe) and run for 15 s:
  • =1: the libc lines read [+0.000000 s] [src/HostCpu.cpp:193 ...] and [+0.011164 s] [src/Thread.cpp:652 RunThread] .... The first AgcDriver line then reads [+0.000000 s] [src/VulkanDevice.cpp:1003 VulkanDevice] ..., followed by [+0.157192 s] ... Vulkan instance created.
  • Unset: no prefix on any line.
• APS5_LOG_TIMESTAMPS across shared libraries, with two libraries that each include LogMacros.hpp and are called 200 ms apart:
  • MinGW GCC 15.2.0 DLLs: the second library's first stamp is 0 us.
  • Linux (g++ 16.1.1, default visibility like the libc and AgcDriver targets): the second library's first stamp is 200109 us, because LogTimestampMicros()::started is an OBJECT UNIQUE symbol shared by both. So the row names the per-library clock as Windows behaviour.
• ANYPS5_LANGUAGE and APS5_LABEL_FLUSH_US change no visible behaviour; their rows were checked against the code and python tools/env_reference.py --table (Source SystemService.hpp, default 250 from the fallback literal). guest_system_service_language passes.
• python tools/env_reference.py --check: the same output as on main, only APS5_NO_WGI ... has no row (documented by docs(env): document APS5_NO_WGI #8657).
• python tools/check_conventions.py --base origin/main: All conventions checks passed.
• ctest --test-dir build --output-on-failure --timeout 120 --parallel 2 -R "^(env_reference|conventions_env_reference|windows_crash_stack_words|windows_crash_without_stack_words|windows_backtrace_report|guest_system_service_language|patched_libraries)$": 7/7 passed, 357 patched libraries loaded, 0 failed.
• python tools/pr_overlap.py --branch agent/envdocs2: conflicts with fix(tools): read the EnvironmentNumber environment reader in env_reference #8995 at L534, the row below APS5_CRASH_INDIRECT_MEMORY, where fix(tools): read the EnvironmentNumber environment reader in env_reference #8995 adds APS5_CRASH_STACK_BYTES. Either order resolves by keeping both rows. Twelve PRs that change ENVIRONMENT.md could not be compared because they conflict with main (feat(agc): trace shader failures through draw submission batches #5431, fix(agc): keep CPU stores made after a deferred DCC key store was recorded #6186, feat(wasm): run relinked titles in the browser with WebGPU #6879, perf(recompiler): keep runtime-specialized shader modules in the shader disk cache #6924, perf(agc): draw finisher pipeline (APS5_DRAW_PIPELINE) #6946, perf(agc): GPU descriptor heap (APS5_GPU_HEAP) #6947, perf(recompiler): leave the descriptor-range comparison out of direct buffer loads when the device zeroes past the range #7484, feat(agc): trace ranges that leave the write watch #7604, feat(agc): look up table images at run time through a device image table on the GPU heap (APS5_GPU_PATH) #7642, fix(shader): split descriptor webs whose paths fold or hold invalid arms #8077, fix(agc): preserve CPU writes after resident linear target draws #8246, perf(agc): expand single-key DCC storage uploads on the GPU from the host import #8484). None of their diffs touch these four rows or the rows next to them.

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: n/a
• AI-assisted: yes, some AI assistance was used

[github-actions]

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 5431	docs/dev/ENVIRONMENT.md
PR 6186	docs/dev/ENVIRONMENT.md
PR 6879	docs/dev/ENVIRONMENT.md
PR 6924	docs/dev/ENVIRONMENT.md
PR 6946	docs/dev/ENVIRONMENT.md
PR 6947	docs/dev/ENVIRONMENT.md
PR 7484	docs/dev/ENVIRONMENT.md
PR 7604	docs/dev/ENVIRONMENT.md
PR 7642	docs/dev/ENVIRONMENT.md
PR 8077	docs/dev/ENVIRONMENT.md
PR 8484	docs/dev/ENVIRONMENT.md

[harisibrar1]

Checked against main:

• APS5_CRASH_INDIRECT_MEMORY: CrashReport.cpp:396-413 always prints the [reg] word lines; GetEnvironmentVariableA(..., nullptr, 0) > 1 turns on [[reg+off]] @addr lines of 64 bytes (content[8]) per pointer, so "off when unset or empty" is right and agrees with USAGE.md:349.
• APS5_LOG_TIMESTAMPS: LogMacros.hpp:26-28 prints [+%llu.%06llu s] , seconds with six microsecond digits; started is a function-local static in an inline function (:19-24), initialised at the first stamped line, so per-library clocks on Windows follow.
• ANYPS5_LANGUAGE: the only reader is SystemServiceConsoleLanguage at SystemService.hpp:41; empty or unset gives English US (1), and tools/env_reference.py --table agrees with the new Source.
• APS5_LABEL_FLUSH_US: DeferredLabels.cpp:60-64 gives 1000 with LabelBatchSubmit() and 250 without; LabelBatchSubmit() (:26-29) is off for APS5_NO_LABEL_BATCH_SUBMIT or APS5_NO_LABEL_BATCHING, so a blank Default with both values in the description is accurate.
• env_reference.py --check on the PR's file reports only the existing APS5_NO_WGI gap, and check_conventions.py --base main --head pr-9036 passes.

Reviewed with AI assistance (Claude Code); the checks above were run locally on my machine.