[savvaniss]

What

This is the follow-up to #8162 that @oneandonlydean asked for in this comment.

Since #8162, a --module-dir module that the executable needs is deferred on Windows. The entry stub maps it at startup, and DynamicLoader runs its module_start and initializers only when the title loads it with sceKernelLoadStartModule. If the executable imports from such a module but the title never loads it, the module gets neither, and nothing reported that.

Now the Windows entry stub prints one line at startup for each deferred guest module that an import of the executable names. The lines come after the imports are bound and before Transferring control to ELF entry point:

Deferred module app0/Media/Modules/needed.prx.guest.prx: the executable imports from it, but it starts only when the title loads it

• Which modules: a guest module with DeferInitialization that is in the search order of one of the executable's own imports, matched by file name or module identity. The stub uses the same match to bind the import. A guest module's own imports do not count. An import with no module name searches every library, so it is not reported.
• What changes: only the executable's entry stub. Guest module bytes stay identical for every executable, so the uniform deferral and --executable conversions are unchanged.
• When it prints: every time such a module is imported, including when the title does load it later, as a Unity player loads mono-ps4.sprx. I left out the optional second line (the first imported call runs before the module has started), because it needs a check on every call through those imports.
• Docs: the debt line from fix(relinker): start Windows runtime modules the executable needs when the title loads them #8162 now says what is reported, instead of "nothing reports it".
• Other targets: Linux and macOS output is unchanged. Since fix(loader): defer Linux runtime module starts until requested #6586 merged, the debt entry lists only macOS as starting such a module at startup; this PR reports only from the Windows entry stub.

Tested

Windows 11 Pro 26300, WinLibs GCC 15.2.0, Release, relinker-only build (-DANYPS5_RELINKER_ONLY=ON -DBUILD_TESTING=ON) on main cecb540a63, then again after merging main 6cd36e6517.

• Full ctest: 233/233 pass on cecb540a63. After the merge, 240/241 pass with -j8; conventions_env_reference timed out at 60 s under load and passes alone in 32 s.
• guest_module_dir (new case): an executable that calls shared from needed.prx.
  • needed.prx in Media/Modules, converted with --module-dir Media/Modules: the Windows executable contains the line, prints it on stdout and exits with 42.
  • needed.prx in sce_module, which starts at startup: the executable contains no such line and exits with 42.
  • Linux: the module is a startup DT_NEEDED entry in both directories.
• Negative control: with main's WindowsEntryStubBuilder.cpp, the case fails (AssertionError: Media/Modules); with the change, it passes.
• Not run on a title. On main alone, Boggle's conversion stops at libSceFios2.prx (the fix is in feat: run PS4 games on the AGC driver — Limbo (CUSA01664) in game (integration branch) #6977).
• tools/check_conventions.py --base origin/main passes.
• tools/pr_overlap.py:
  • fix(loader): defer Linux runtime module starts until requested #6586 rewrote the same debt sentence and has merged; the merge keeps its wording and this PR's sentence. fix(windows): enter guest startup on an arena-backed main stack #3965 adds code at the same place in the entry stub, just before guestStartup.Initialize, which would only be a textual conflict.
  • feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021, feat: run PS4 games on the AGC driver — Limbo (CUSA01664) in game (integration branch) #6977, fix(relinker): rewrite XSAVEOPT to XSAVE under --to-rosetta #7918 and fix(relinker): mask embedded data tables in inter-FDE gaps #8076 edit other entries of docs/dev/debt/relinker.md. fix(relinker): share Windows --module-dir modules between executables #7450, which touched test_guest_module_dir.py, is closed.

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

🤖 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 3965	core/relinker/elfpatcher/src/windows/WindowsEntryStubBuilder.cpp L719

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/dev/debt/relinker.md
PR 7173	docs/dev/debt/relinker.md
PR 8076	docs/dev/debt/relinker.md

[BKlaasWerkman]

Rechecked on current main: head 960863f merged into main 990f45e (clean merge). Windows 11 Pro 10.0.26300, WinLibs GCC 15.2.0 r7, relinker-only BUILD_TESTING=ON, Release: 235 of 235 pass, including guest_module_dir with the new case, and none changes result against a run of that main on the same machine.

#10157 also changes WindowsEntryStubBuilder.cpp, after SetStdHandle. The two PRs merge cleanly with each other on top of that main; I'll post a build and test of the combination once it has run.

AI-assisted: yes (Claude Code)

[BKlaasWerkman]

The combination, as promised: main 990f45e + this PR (960863f) + #10157 (1e79db7), merged in that order with no conflict. Same machine and toolchain (Windows 11 Pro 10.0.26300, WinLibs GCC 15.2.0 r7, relinker-only BUILD_TESTING=ON, Release). It builds, and the full suite passes 235 of 235, with no result changed against that main. That includes guest_module_dir with this PR's new case and windows_import_modules with #10157's suspended-process preload case. The two entry-stub changes work together in either merge order as far as these tests reach.

AI-assisted: yes (Claude Code)