perf(libc): set TZ once on Linux

[Aryansh1520]

What

Without TZ, glibc's localtime and mktime run tzset on every call, and each tzset re-checks /etc/localtime. PPSA28997 (God of War Sons of Sparta) calls them every frame during its intro, which then spent most of its time there and played back far below speed. libc now sets TZ to :/etc/localtime when it is unset and calls tzset once at load, so the zone file is read once. A TZ set by the user is left as it is. Windows is unchanged.

Tested

• Linux (CachyOS, kernel 7.2), GCC: PPSA28997's intro plays at normal speed; local times are unchanged.
• tools/check_conventions.py --base origin/main passes.

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
• AI-assisted: yes

[MarcZX10]

The optimization and evidence are Linux/glibc-specific, but #ifndef _WIN32 also runs this initializer on macOS. That changes the process environment there and forces an explicit /etc/localtime zone-file path without Darwin validation. Could you gate this with __linux__ (or provide a macOS behavior test and evidence) so the change only affects the platform described in the PR? I’m holding approval pending that.

[Aryansh1520]

Agreed. The initializer is now behind #ifdef __linux__, so macOS and Windows are unchanged and the change only affects the platform the PR describes.

[MarcZX10]

Thanks for limiting the initializer to __linux__. I built exact head 631654b6 on Linux with GCC 12 and ran guest_time_conversion with TZ unset; it passed 1/1 in 0.46 s. Conventions and git diff --check pass. GitHub currently reports no checks on this head, so please get the required CI jobs running (or have a maintainer approve any held workflow) before merge. I will hold approval until checks are visible.

[oneandonlydean]

Could you add numbers for the PPSA28997 intro, main against main plus this (fps or frame time over a few runs)? Perf changes need before/after figures for the change itself (#218).

[Aryansh1520]

PPSA28997, Linux (CachyOS, kernel 7.2), GCC, same build with and without this change: the time from the Sony logo to the Santa Monica logo drops from about 2 minutes to about 15–20 seconds, which is normal speed. Without it, the intro plays far below speed the whole time because localtime/mktime re-read the zone each frame.

[popcoder565]

Native Windows run, since no Windows results were posted. Setup: Windows 10 (19045), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release. Head 631654b6, merged onto main d70b899 in one build with 38 other PRs that had no Windows run.

• The change is Linux-only, so on Windows this only checks that it builds.
• Full ctest in that build: 519 of 519. That needed local fixes to three other PRs in the batch that don't build or pass on Windows: feat(libc): export scriptingGetMem and scriptingFreeMem from application heap #1611, feat(posix): implement gethostbyname #1796 and feat(zlib): implement sequential libSceZlib decompression #1824 (reported there).

No title was run.

🤖 Addressed by Claude Code

[adriadam10]

Checked the diff against glibc (tzset_internal has no early return when TZ is unset, so every localtime/mktime re-reads the zone file), built the libc tests on Linux and ran them (42/43; guest_time_conversion passes, the one not run is guest_wcsrtombs, whose executable isn't in the scoped build). Numbers for PPSA28997 are in the thread, Linux-only gating is in, and Windows builds per @popcoder565. Merging.