[simonbronk]

What

Fixes #4497.

Weak undefined imports (STB_WEAK, section 0) in the input ELF were written to the relinked .dynsym as STB_GLOBAL, so a title that weak-imports a function no module exports stopped at startup instead of binding the import to null.

• NidReference records whether the source symbol was weak; SysVDynamicSectionBuilder writes STB_WEAK for those imports instead of STB_GLOBAL. On Linux, ld.so then binds a missing weak import to 0.
• Windows: WindowsRelocationBuilder reads the binding into PeImport::Weak. When no searched module exports a weak import, the startup stub writes 0 plus the addend to its slot and continues; it does not print unresolved ELF import or install a lazy stub. A weak import that is exported resolves as before.
• Non-weak imports are unchanged: a missing one still fails at startup.

Guest module imports (LinuxGuestModuleWriter, WindowsGuestModuleWriter) are not touched.

Tested

macOS 27 (Apple Silicon, AppleClang 21), relinker-only build, Release:

• ctest --test-dir build-relinker: 125/125 passed, including the new weak_imports.
• weak_imports builds an executable with one weak GLOB_DAT import whose code returns 33 when the slot is null and calls through it otherwise:
  • Linux output: the import is STB_WEAK, and a non-weak import is still STB_GLOBAL. Without the fix the test fails with {'absent': 1}.
  • Windows output: relinks with the import missing and present; on Windows it runs and expects 33 (missing) and 11 (exported by a.prx). I could not run the Windows executables; this part needs Windows CI.
• python3 tools/check_conventions.py --base origin/main: passed.

Porpoise itself not tested (no dump).

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 in TechnicalDebt
• 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
• Unblocks: PPSA99764 (Porpoise), no Game bring-up discussion found
• AI-assisted: yes (Claude Code, Opus 5.5)

🤖 Generated with Claude Code

[simonbronk]

Overlaps with #4452 on the same lines in SysVDynamicSectionBuilder.cpp and RelinkerPipeline.cpp, but the changes are independent and compatible (#4452 handles all-zero symbol entries, this one keeps STB_WEAK from named symbols). Happy to rebase onto #4452 if it lands first.

The CMakeLists.txt overlap with #4062 is resolved: the test registration moved next to elf_header.

[gabichuuuuu]

Tested the Windows part on Windows.

Setup: Windows 11 Enterprise 10.0.26200 x64, WinLibs GCC 15.2.0 r7 (15.2.0posix-14.0.0-ucrt-r7, as in CI), CMake 4.3.1, Ninja 1.13.2, relinker-only Release build with BUILD_TESTING=ON, Python 3.12.

• This PR (bdf003a5) merged into main 2a5eca11, as the CI merge step does: ctest --test-dir build-relinker --output-on-failure --timeout 120 passed 130/130, including weak_imports. main alone passed 129/129; no other test changed result.
• weak_imports runs the converted executables here: windows-weak-missing exits 33, windows-weak-present exits 11.
• Same Windows cases with the main relinker: windows-weak-missing stops with 0xc0000139 (FAIL: unresolved ELF import absent from a.prx, GetLastError: 127); windows-weak-present exits 11. The Linux part of the test fails on main with ('linux-weak', {'absent': 1}), as described above.

Review note, no change requested: the weak branch in WindowsEntryStubBuilder skips the RelocationType 16/17 loads that the non-weak resolved branch does. It is unreachable now, since only executable imports (type 1) set Weak and guest module imports keep Weak = false.

[simonbronk]

@gabichuuuuu thanks for running it on Windows.

#4452 is merged, so I merged main into this branch: symbolInfo from #4452 is kept (all-zero entries stay STB_WEAK | STT_NOTYPE) and named imports now take STB_WEAK or STB_GLOBAL from the source symbol. Relinker-only build on macOS: 134/134 tests pass; conventions check passes.

On the review note: agreed, the 16/17 loads are unreachable in the weak branch because only executable imports set Weak, so I left it as is.