[hypermoist]

What

lgamma returned the host's value but never stored the sign of Γ(x), because libc exported no signgam. The debt entry in docs/dev/debt/libc.md recorded this.

libc now exports signgam (NID 5gOOC0kzW0c, from the aerolib symbol list), starting at 0 as in FreeBSD 11's s_signgam.c. lgamma sets it by the rules of FreeBSD 11's e_lgamma_r.c:

• ±0 gives the sign of the zero;
• infinities and NaN give 1;
• negative integers (poles) give 1. glibc differs here and reports a parity-based sign;
• a negative non-integer gives -1 when its floor is odd (the sign of sin_pi(x)), 1 otherwise.

The returned value is unchanged; it is still the host's lgamma.

Overlap: #5775 rewrites the same lgamma line, to handle subnormal arguments on Windows. Whichever merges second needs a trivial rebase that keeps both the subnormal branch and the signgam store.

Tested

CachyOS Linux (kernel 7.2.9), GCC 16.2.1, Release, on main 8856f18.

• guest_math checks the stored sign for positive values, ±0, -0.5, -1.5, -2.5, a tiny negative, negative integers including -2^60, a value just above -2^52, ±inf and NaN.
• A separate host program compared the rule with glibc's lgamma_r over 2,000,000 values in [-200, 200], including integers and values scaled by 2^-70. There were 0 mismatches outside the negative-integer poles, where glibc and FreeBSD intentionally differ.
• Pass: guest_math. tools/check_conventions.py --base origin/main passes.
• Windows: not built or tested.

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
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

🤖 Generated with Claude Code

[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/dev/debt/libc.md
PR 5303	docs/dev/debt/libc.md
PR 6044	core/libs/tests/GuestMath.cpp
PR 6061	docs/dev/debt/libc.md
PR 6154	core/libs/prx/libc/src/Math.cpp core/libs/tests/GuestMath.cpp
PR 6157	core/libs/prx/libc/src/Math.cpp core/libs/tests/GuestMath.cpp
PR 6158	core/libs/prx/libc/src/Math.cpp core/libs/tests/GuestMath.cpp
PR 6161	core/libs/prx/libc/src/Math.cpp core/libs/tests/GuestMath.cpp
PR 6163	core/libs/prx/libc/src/Math.cpp core/libs/tests/GuestMath.cpp
PR 6222	docs/dev/debt/libc.md
PR 6414	core/libs/prx/libc/src/Math.cpp
PR 6565	docs/dev/debt/libc.md

[hypermoist]

Rebased onto current main (5d9dcc4). The conflict was with the new tiny-argument path in lgamma (-log|x| below 2^-56): signgam is now set first and that path is kept unchanged, and both the new subnormal tests and the signgam tests stay in GuestMath.cpp. guest_math passes and check_conventions.py is clean.

🤖 Generated with Claude Code

[harmsluis]

Windows run (Windows 11 build 26300, non-elevated, WinLibs MinGW-w64 GCC 15.2.0 r7, Ninja, Release, BUILD_TESTING=ON), built target guest_math_tests:

• main c84249186 (baseline): guest_math passes.
• This PR (20edd079e) merged onto c84249186 without conflicts: builds with no new warnings, and guest_math passes 5/5 repeated runs (ctest --repeat until-fail:5).
• The unpatched libc.prx exports signgam_nid_postfix as data (objdump -p). The test sets the variable to 7 before each lgamma call and reads it through that import, so the sign checks exercise the DLL's variable rather than a local copy.

Not covered: the full build, and the patched/NID-renamed libc.prx.