[UnhealthyKraken]

What

Implement the missing feraiseexcept export (NIfFNcyeCTo) by accumulating requested flags in both x87 and MXCSR, preserving other state and errno, and delivering enabled x87 exceptions with fwait. All six defined exception bits are supported; unverified bits throw before changing state. This deliberately does not call _Feraise, whose errno and range/inexact behavior differs.

The signature is from the public SDK fenv.h. The public 3.20 table and symbol database corroborate the NID. For the six supported bits, behavior follows FreeBSD amd64 fenv.c; console equivalence and module/library versions remain unmeasured and are recorded in debt.

Unknown bits are explicitly rejected before changing state, unlike the FreeBSD reference, which does not mask or validate the requested bits. This rejection is an unsupported-path policy; PS5 behavior for those bits is unknown.

Tested

Windows x86-64, GCC15.2 Release, isolated checkout based on 1ff5c8ac5c3050d71360957c5ff1be38aaac7114. Regression first: seven new tests fail baseline's missing export while the existing floating-point test passes. With the same final executables and only the unpatched libc swapped, baseline is 1 pass/7 fail; candidate is 8/8 pass.

The main case checks 196608 flag/control combinations: all 64 requested and prior flag sets, independent x87/SSE flags, 16 rounding combinations and three x87 precision modes. It checks errno, other control/status/tag/pointer fields, rejected masks, differing exception masks, and thread isolation. Six separate cases check enabled x87 traps; Windows validates the saved unmasked cause as well as the exception code. A direct native x87 load of the smallest subnormal and the implementation both produce Windows 0xc0000090 on this host, so the denormal case accepts that code only when the saved active flag is DE, not invalid.

At the original submitted head 2d22f777, the full Windows libs build succeeded; validation was 9/9 pass, including all 354 patched libraries loading with zero failures. The expected NID is verified in patched shared libc, which supplies the interface for libSceLibcInternal under the existing provider convention. Loading verifies linkage only. No Linux/macOS or console execution by this contributor. The POSIX trap handler confirms SIGFPE but does not inspect its saved cause. No games run.

Conventions, whitespace and registration checks pass (1148 targets, none unregistered). Independent read-only review found no actionable defect and did not rerun tests. The overlap report's 14 base-conflicted shared-file warnings were reviewed manually: no competing export. #6044 reorganizes tests; #6879 changes host math callers rather than adding this export. The original overlap run had no compared-file conflict or duplicate export. The review-correction scan also has none; its 14 current base-conflicted shared-file warnings were inspected at their fetched heads and concern unrelated stack/runtime, formatting, environment, archive, directory and test-reorganization work. #6044 remains a test-directory migration consideration when rebased. The debt paragraph was relocated away from #8397/#8287 documentation changes.

Review correction: the thread test now sets its errno sentinel after the first host barrier and captures errno immediately after the export, before the second barrier. The floating-point isolation checks and distinct thread sentinels are retained; no runtime code changed. Filip722's Linux review reports 22/30 original runs passing, barrier-induced errno changes, and 30/30 passing with this test correction; those are attributed reviewer results, not execution on this Windows host.

Locally, a deterministic temporary probe makes both barriers set errno to 789: the original test fails and the repaired test passes against the same preserved DLL. A negative control that changes errno inside the called export still fails the repaired assertion. The rebuilt eight Windows tests pass, the main test passes 30 consecutive runs, and identical final executables pass 8/8 against both preserved baseline and candidate DLLs (byte-identical SHA256 bf4a81f2a272152f4886821d3c432b6ce14b98302037c5439a91297888ad8a75). These review checks rebuild the affected targets, not the entire library collection. Conventions and whitespace checks pass; independent read-only review found no actionable issue and did not rerun tests.

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

[github-actions]

Progress report (cceb6c3 → 676ee4f)

➖ System libraries: 50.06% (+0.0%, +1 implemented functions)

✅ 1 implemented

Library	Name
libc	feraiseexcept_nid_postfix

Changed libraries (implemented exports)

Library	Before	After	Change
libc	1347/1357 (99.26%)	1348/1358 (99.26%)	+0.00 pp

[MarcZX10]

I ran an additional native Linux check on exact head 2d22f777eb with GCC 12.2. I compiled the PR’s FloatingPoint.cpp into a temporary shared object and its existing GuestRaiseExcept.cpp into a host harness, supplying only a temporary __error_nid_postfix stub for this isolated test. The normal state/errno/thread cases pass, and all six enabled-x87 trap invocations exit through the test’s SIGFPE handler. The PR conventions checker and git diff --check also pass. This validates the implementation on this Linux host; it does not validate the built AnyPS5 libc or inspect the POSIX trap cause.

[MarcZX10]

Reviewed the new export and regression coverage. The implementation updates x87 and MXCSR flags independently, rejects unsupported bits before changing state, preserves errno/control fields, and uses fwait for enabled x87 exceptions. I independently ran the PR test on Linux with GCC 12.2 through a temporary host harness; the state, errno, thread, and six trap cases passed. Console behavior remains unmeasured and is clearly recorded in debt. The platform jobs still running are not covered by this approval.

[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	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 5303	docs/dev/debt/libc.md
PR 6713	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7081	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7201	core/libs/prx/libc/CMakeLists.txt
PR 7386	docs/dev/debt/libc.md
PR 7447	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7544	docs/dev/debt/libc.md
PR 7648	docs/dev/debt/libc.md
PR 7715	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7851	core/libs/prx/libc/CMakeLists.txt core/libs/prx/libc/include/Api.hpp
PR 8150	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 9371	docs/dev/debt/libc.md
PR 9447	docs/dev/debt/libc.md
PR 9526	docs/dev/debt/libc.md
PR 9775	core/libs/prx/libc/CMakeLists.txt
PR 9929	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 9953	docs/dev/debt/libc.md
PR 9964	docs/dev/debt/libc.md
PR 10002	docs/dev/debt/libc.md
PR 10007	docs/dev/debt/libc.md
PR 10158	docs/dev/debt/libc.md
PR 10200	docs/dev/debt/libc.md
PR 10234	docs/dev/debt/libc.md
PR 10315	docs/dev/debt/libc.md
PR 10317	docs/dev/debt/libc.md
PR 10321	docs/dev/debt/libc.md
PR 10325	docs/dev/debt/libc.md
PR 10328	docs/dev/debt/libc.md
PR 10345	docs/dev/debt/libc.md
PR 10349	docs/dev/debt/libc.md
PR 10405	docs/dev/debt/libc.md
PR 10467	docs/dev/debt/libc.md
PR 10506	docs/dev/debt/libc.md
PR 10510	docs/dev/debt/libc.md
PR 10512	docs/dev/debt/libc.md
PR 10529	docs/dev/debt/libc.md
PR 10565	docs/dev/debt/libc.md
PR 10749	core/libs/prx/libc/include/Api.hpp docs/dev/debt/libc.md
PR 11332	docs/dev/debt/libc.md
PR 11341	docs/dev/debt/libc.md

[r1p0N]

The implementation matches FreeBSD's amd64 feraiseexcept (fesetexceptflag on x87 status and MXCSR, then __fwait()), so valid masks behave the same way.
One difference worth stating in the debt entry: FreeBSD neither masks nor validates excepts, while this throws std::invalid_argument for bits outside the six exceptions. That is consistent with the throw-on-unknown rule and avoids writing reserved MXCSR bits, but it is a deliberate deviation from the console's likely behaviour, and the debt note should say so if it does not already.
Not run: the tests (they dlopen the built libc and need the full build).

[Filip722]

Native Linux full-project validation on exact head 2d22f777eb4b4928df17f0aeeb211f61ae2d70f9 found an intermittent issue in the new test's Threads() errno assertion. The export itself preserved errno in the instrumented runs; the host barriers changed it outside the call.

Environment: Debian 13.7 x86-64, kernel 7.2.9-x64v3-xanmod1, Ryzen 5 7640HS, GCC/libstdc++ 15.2.0, CMake 3.31.6, Release, bundled dependencies, BUILD_TESTING=ON, ANYPS5_RELINKER_ONLY=OFF, two build jobs. Built the actual libc, guest_raise_except_tests, guest_floating_point_tests, and nid_patcher; no stubbed errno provider.

Repro / diagnosis:

• The unmodified guest_raise_except executable passed 22/30 runs and aborted 8/30 with feraiseexcept: thread-local errno. The focused CTest run also reproduced it.
• A temporary diagnostic copy read errno immediately before the export, immediately after it, and after the second ready.arrive_and_wait(). Failing examples were flag=4 before=104 after=104 afterBarrier=11 and flag=1 before=11 after=11 afterBarrier=11. Thus either host barrier can spoil the sentinel; the export preserves the value it receives. The real project's __error_nid_postfix() returns &errno; this libstdc++ uses futex waits in the barrier's atomic-wait path and allows EAGAIN (11) without restoring errno.
• A minimal temporary correction passed 30/30 runs against the same built libc: set the sentinel after the first barrier and capture errno before the second, then assert the captured value. The existing floating-point environment checks can stay after the barrier:

ready.arrive_and_wait();
*__error_nid_postfix()=static_cast<int>(100+flag);
Require(raiseFlags(static_cast<int>(flag))==0,"thread return");
const int raisedError=*__error_nid_postfix();
ready.arrive_and_wait();
// Keep the existing environment checks here, then:
Require(raisedError==static_cast<int>(100+flag),"thread-local errno");

Additional passing evidence:

• All six original enabled-x87 trap cases and the existing guest_floating_point control passed three consecutive runs each.
• Extended a temporary copy of the existing POSIX test with sigaction(SA_SIGINFO) and checked SIGFPE plus uc_mcontext.fpregs->swd & ~uc_mcontext.fpregs->cwd & 0x3f against exactly the requested bit. All six bits, including denormal, passed three runs each. This checks the saved active cause in addition to signal delivery.
• Ran the project's NID patcher on a temporary copy of the actual libc; nm -D --defined-only contains exactly one NIfFNcyeCTo export. This is packaging verification, not a full patched-runtime/title test.
• Negative control: restored only FloatingPoint.cpp from merge-base 1ff5c8ac5c3050d71360957c5ff1be38aaac7114 and rebuilt libc, retaining the final test executables. The seven new cases failed with the missing-export assertion; the existing floating-point case passed. After restoring the exact PR implementation, the six trap cases and old control passed again, while the original main case reproduced the barrier/errno failure. The corrected temporary test and saved-cause checks passed against the restored library.

Build/repro commands:

CCACHE_DIR=/tmp/anyps5-ccache cmake --build build --target libc guest_raise_except_tests guest_floating_point_tests nid_patcher --parallel 2
ctest --test-dir build --output-on-failure --parallel 2 --repeat until-fail:3 -R '^(guest_raise_except(_trap_[0-5])?|guest_floating_point)$'

All diagnostic variants stayed under /tmp; repository sources are restored to the exact head. No Windows, console or game execution in this check. AI-assisted testing/reporting: yes.

[UnhealthyKraken]

Addressed both review points in 35c508e.

The thread test now sets its distinct errno sentinel after barrier one and captures it immediately after feraiseexcept, before barrier two. The floating-point checks remain unchanged. A deterministic barrier-side errno injection fails the old test and passes the repaired test using the same DLL; an export-side errno mutation still fails the repaired assertion. On Windows, all eight relevant tests pass, the main test passes 30 consecutive runs, and identical final executables pass 8/8 against both preserved and candidate DLLs. The DLLs are byte-identical; no runtime code changed. Filip722's Linux results remain attributed reviewer evidence; I have not run Linux here.

The debt entry now explicitly states that FreeBSD does not mask or validate the requested exception bits, whereas this implementation rejects unknown bits before modifying state. Console behavior for those bits remains unknown. Conventions and whitespace checks pass. The PR description separates the original full-library validation from these focused review checks.

[oneandonlydean]

@UnhealthyKraken #10688 landed and conflicted with this in docs/dev/debt/libc.md, so I merged main into the branch (676ee4f). The only conflict was the two neighbouring debt bullets, and both are kept; no code changed. Next is CI on this head.

[UnhealthyKraken]

@oneandonlydean Recorded your history-preserving main merge 676ee4fe427ff683637d5725bce6b4acf2c18be3, including both neighboring debt entries. The applicable hosted checks on this head have now completed successfully; full product Linux/Windows jobs are skipped. No additional source change is needed for the conflict you resolved, and I have not rewritten your merge or claimed a new runtime run. No local builds/tests. AI-assisted: yes.