[dantdmdarealguy]

What

Since #6067 the Windows libs target also writes mimalloc.dll and mimalloc-redirect.dll to build/core/libs/libs (core/libs/CMakeLists.txt L207-227). tools/package_release.py L55 puts both in the release archives. The Windows entry stub loads mimalloc.dll from the run path only when APS5_MIMALLOC is set (WindowsEntryStubBuilder.cpp L103-133, ARCHITECTURE.md L65-66). The runtime layout in USAGE.md tells a self-built libs/ to take build/core/libs/libs/*.prx and the three MinGW runtime DLLs, and nothing else. #6067 did not change USAGE.md. A libs/ set up that way runs with APS5_MIMALLOC set, but it never loads mimalloc and prints nothing.

The paragraph on the MinGW runtime DLLs now names the two mimalloc DLLs: they are loaded from libs/ with APS5_MIMALLOC, they are copied from build/core/libs/libs, and without them the title keeps the C runtime's heap and prints nothing.

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 (15.2.0posix-14.0.0-ucrt-r7), Git for Windows 2.51.0.

Setup: a fresh clone of main (6696def6e), with the Windows steps of docs/dev/BUILD.md and docs/user/USAGE.md run in the order given, in Windows PowerShell:

• git submodule update --init --recursive
• the $env:PATH line, with C:\winlibs\mingw64\bin as the toolchain path
• cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++, with SSL_CERT_FILE set to Git's mingw64 CA bundle as the Requirements say
• cmake --build build
• cmake --build build --target libs
• the runtime DLL Copy-Item loop

ScreenTester5 (PPSA69420) was then unwrapped with tools/unwrap_self.py, relinked with relinker --windows source/eboot.elf run/app.exe and laid out as the runtime layout says: libs/ with build/core/libs/libs/*.prx and the three MinGW DLLs, and app0/ with eboot.bin, sce_sys/ and the converted sce_module/. .\app.exe runs to its menu.

• Before, with the layout as documented: $env:APS5_MIMALLOC = "1"; .\app.exe. The title runs, but no mimalloc* module is loaded in the process ((Get-Process -Id <pid>).Modules), and stdout and stderr have no line about it. With only mimalloc.dll copied, nothing is loaded either.
• After, with both DLLs copied from build/core/libs/libs into libs/ as the changed paragraph says: the same command loads libs\mimalloc.dll and libs\mimalloc-redirect.dll, and the title runs to its menu.
• python3 tools/check_conventions.py --base origin/main passes. tools/pr_overlap.py --branch agent/docsweep reports no overlap with the PRs it could compare. It could not compare 9 PRs that conflict with main (feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021, perf(agc): build the pipelines of a submission ahead on a compile pool #5925, feat(agc): select Vulkan devices by driver name #6788, 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, feat(agc): look up table images at run time through a device image table on the GPU heap (APS5_GPU_PATH) #7642, fix(agc): reclaim idle images safely during cache creation #7800, fix(relinker): rewrite XSAVEOPT to XSAVE under --to-rosetta #7918). None of their USAGE.md hunks touch this paragraph or mention mimalloc.

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

What was run, step by step (fresh clone X:/aps5-agents/docsweep-work/AnyPS5, main 6696def6e)

Documented step	Result	Adjusted
BUILD.md "Full build": git submodule update --init --recursive	ok (2 min 45 s)	none (clone used --reference-if-able to the local clone and submodule.alternateLocation=superproject to save bandwidth; this does not change the commands)
BUILD.md Requirements: WinLibs 15.2.0posix-14.0.0-ucrt-r7, Git with Bash, Python 3	present	WinLibs at C:\winlibs\mingw64
BUILD.md Windows setup: $env:PATH = "...\mingw64\bin;C:\Program Files\Git\cmd;C:\Program Files\Git\usr\bin;$env:PATH"	ok	toolchain path C:\winlibs\mingw64\bin instead of C:\tools\winlibs\... (the doc says to adjust)
BUILD.md Commands: cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++	without SSL_CERT_FILE: Could not download FFmpeg 9ac4cfd195f1: 60;"SSL peer certificate or SSH remote key was not OK...", exactly as the Requirements bullet predicts. With $env:SSL_CERT_FILE = "C:\Program Files\Git\mingw64\etc\ssl\certs\ca-bundle.crt" (Git 2.51 < 2.56, so mingw64, as the doc says): ok	none
cmake --build build --parallel	ok, 487 steps, no .prx built (matches "With the default BUILD_TESTING=OFF, the default build does not produce them")	--parallel 6 (machine rule)
cmake --build build --target libs --parallel	ok, 1835 steps, 354 .prx in build/core/libs/libs plus mimalloc.dll, mimalloc-redirect.dll, unpatched/, implib/. 6 warnings, all in glslang or libstdc++ headers	--parallel 6
Runtime DLL Copy-Item loop	ok	$toolchainBin = "C:\winlibs\mingw64\bin"
USAGE.md: tools/unwrap_self.py on the title's eboot.bin (SELF 4f153d1d) and sce_module/libc.prx (fSELF 5414f5ee)	ok	input copied from X:/aps5-agents/homebrew1-titles/PPSA69420-hb1/extracted/PPSA69420
USAGE.md: relinker --windows source/eboot.elf run/app.exe	ok: 208 NID references, app0/sce_module/libc.prx.guest.prx	relinker is not on PATH; ran build/core/relinker/relinker.exe (see follow-up 1)
USAGE.md runtime layout + .\app.exe (SDL_AUDIODRIVER=dummy)	ScreenTester5 runs to its menu on the RTX 5080 (TV, screenshot X:/aps5-agents/docsweep-work/shots/screentester5.png), no error lines	none

README.md has no Windows steps; it links to BUILD.md and USAGE.md. Nothing in it to follow or fix.

All the main-path steps work as written. The only verified inaccuracy is the one fixed: the Windows runtime layout leaves out the two mimalloc DLLs.

Evidence for the fix

• core/libs/CMakeLists.txt L207-227: if(TARGET mimalloc) copies mimalloc.dll (with pe_import_first) and mimalloc-redirect.dll into PATCHED_DIR (build/core/libs/libs) as libs outputs. The mimalloc target is created in the root CMakeLists.txt L358-363 (the MinGW branch).
• WindowsEntryStubBuilder.cpp L103 and L125-133: APS5_MIMALLOC and path + "mimalloc.dll", where path is the run path (libs/ by default). ARCHITECTURE.md L65-66 says the same and says the load fails quietly without the DLL.
• tools/package_release.py L54-55: the Windows archive gets the three MinGW DLLs and both mimalloc DLLs. The release_package ctest passes.
• git show --stat 9824c46fe (feat(libs): let relinked Windows games opt into mimalloc #6067, merged 2026-10-19) changed ARCHITECTURE.md and ENVIRONMENT.md but not USAGE.md.
• Measured with ScreenTester5 and APS5_MIMALLOC=1, reading (Get-Process).Modules 12 s after launch (X:/aps5-agents/docsweep-scripts/mimalloc_probe.ps1):
  • documented layout: none loaded;
  • mimalloc.dll only: none loaded;
  • both DLLs: libs\mimalloc.dll and libs\mimalloc-redirect.dll loaded.
  • In all three runs stdout and stderr have no mimalloc text, and the title reaches its menu.
• No open or closed PR or issue covers this (gh pr list --state all --search mimalloc, gh issue list --search mimalloc). Claimed on the board (feat(shader): implement MIMG sample/gather flag variants #226), with finding refactor(ajm): drop duplicated stub exports from libSceAjm and libSceAjmi #228.

Verification of the branch ($BUILD_DIR = X:/aps5-agents/docsweep-build)

• Before configuring, the clone had four uninitialized submodules (ZArchive, mbedtls, mimalloc, zstd). The first configure failed on them. git submodule update --init --recursive fixed it. This was my environment, not a doc problem.
• cmake --build "$BUILD_DIR" --parallel 6 fails on one target only: agc_driver_image_array_window_past_layer_limit_tests (ImageArrayWindowPastLayerLimit.cpp:33:52: error: requested alignment '65536' exceeds object file maximum 8192). It is on unmodified main (from ed599f525), it is MinGW-only and winsweep3 already reported it. Open PR test(agc): align the array window test's texels at run time #8310 fixes it. With -- -k 0 every other target builds.
• cmake --build "$BUILD_DIR" --target libs --parallel 6: ok.
• ctest --test-dir "$BUILD_DIR" --output-on-failure --timeout 120 --parallel 2 -R "^(release_package|patched_libraries|pe_import_first)$": 3/3 pass, 354 patched libraries loaded, 0 failed.
• python tools/check_test_registration.py "$BUILD_DIR": 1140 test executable(s), 0 built and not run as recorded, none new.
• python tools/check_conventions.py --base origin/main: All conventions checks passed.
• The change adds no test: it is documentation only, and the before/after evidence is the module probe above.

Other findings, not fixed (for follow-ups)

1. USAGE.md calls relinker ... as if it were on PATH. BUILD.md names the executable's path only for the relinker-only build (build-relinker/core/relinker/relinker). The full build writes it to build/core/relinker/relinker.exe, and the docs never say so. A reader of the Windows steps has to find it. This is minor and only a missing pointer.
2. BUILD.md's "Windows setup and troubleshooting" section, read alone, does not mention SSL_CERT_FILE. The configure command fails with status 60 unless the reader has applied the Requirements bullet. The doc is correct, but the Windows section's order invites the failure. A one-line $env:SSL_CERT_FILE = ... beside the $env:PATH line would make the sequence runnable as written. I didn't change it, to keep one topic, and because docs(build): note the Group Policy script block on libusb's version step #8568 and docs(build): point the antivirus note at the relinker tests' test-temp #8619 are open on BUILD.md.
3. BUILD.md L90 ("When preparing a title's libs/ directory, copy these DLLs with the patched .prx files") has the same omission as USAGE.md. It could point at the USAGE paragraph. It's a different file, so it was left out of this commit.
4. The relinker's own "Arrange the files like this" message lists only libs/*.prx and app0/..., not the MinGW runtime DLLs that a self-built Windows libs/ needs. That's code, not docs.
5. BUILD.md's CMake flag table leaves out ANYPS5_RELINKER_CLI (core/relinker/CMakeLists.txt:1, default ON; forced OFF for WASM). Not Windows-specific.
6. Verified correct, no change needed:
   • WinLibs release and archive name (same as build.yml/release.yml);
   • CMake 3.22.1 (required by the SPIRV-Tools, Vulkan-Headers and glslang submodules);
   • Python 3 for glslang (StandAlone/CMakeLists.txt:34);
   • libusb gen-describe.ps1 with Bash/cygpath;
   • the libs / patched_libraries_tests dependency;
   • the build/core/libs/libs and unpatched paths;
   • the test names in the antivirus and Bitdefender bullets;
   • the agc_*_replay targets and paths;
   • the Tracy file names;
   • the release archive contents.

Not done or uncertain

• I couldn't check the Git for Windows 2.56 ucrt64 CA-bundle path, because this machine has Git 2.51 (mingw64). release.yml tries both.
• The full ctest suite was not run, only the tests above, because the change is documentation only.
• Nothing was pushed or posted to GitHub. Board posts: feat(libSceKeyboard): implement the keyboard library #225 (status), feat(shader): implement MIMG sample/gather flag variants #226 (claim), feat(libScePngEnc): implement PNG encoding #227 (result: relinked ScreenTester5 at X:/aps5-agents/docsweep-work/title/run), refactor(ajm): drop duplicated stub exports from libSceAjm and libSceAjmi #228 (finding).

[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 5021	docs/user/USAGE.md
PR 5925	docs/user/USAGE.md
PR 6171	docs/user/USAGE.md
PR 6586	docs/user/USAGE.md
PR 6788	docs/user/USAGE.md
PR 6924	docs/user/USAGE.md
PR 7642	docs/user/USAGE.md
PR 7800	docs/user/USAGE.md
PR 7918	docs/user/USAGE.md
PR 8016	docs/user/USAGE.md
PR 9408	docs/user/USAGE.md
PR 9508	docs/user/USAGE.md

[harisibrar1]

Checked against main:

• core/libs/CMakeLists.txt:207-227 writes mimalloc.dll and mimalloc-redirect.dll to build/core/libs/libs as libs outputs, and tools/package_release.py:58 puts both in the Windows archive, so "copy both from build/core/libs/libs; the release archives contain them" matches.
• WindowsEntryStubBuilder.cpp:103,125-133 loads <run path>mimalloc.dll only when APS5_MIMALLOC is set, and docs/dev/ARCHITECTURE.md:65-66 says the load fails quietly without the DLL, so "keeps the C runtime's heap and prints nothing" is accurate.
• docs/user/USAGE.md:169,177 on main lists only the .prx files and the three MinGW DLLs, so the two DLLs were missing from the runtime layout.
• tools/check_conventions.py --base main --head pr-8909 passes.

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