[sktharunbalaji13-cmd]

What

Implement the missing sceKernelFdatasync libkernel export (NID 30Rh4ixbKy4), which is imported by Dying Light 2 (PPSA02262). The implementation uses _commit on Windows and fdatasync on POSIX, following the existing sceKernelFsync implementation pattern.

Add regression coverage for successful synchronization and invalid POSIX file descriptors. The related open PR #1620 implements a distinct fdatasync_nid_postfix export with a different NID, so it does not overlap with this change.

Tested

Windows/MinGW Release:

• cmake --build build --target guest_filesystem_tests — passed.
• ctest --test-dir build -R "^guest_filesystem$" --output-on-failure — 1/1 passed.
• Verified sceKernelFdatasync is exported and the repository NID helper returns 30Rh4ixbKy4.

WSL/Linux Release:

• cmake --build build-libs --target guest_filesystem_tests — passed.

• ctest --test-dir build-libs -R '^guest_filesystem$' --output-on-failure — 1/1 passed.

• Verified the exported symbol; both valid-descriptor and invalid-descriptor assertions passed.

• Confirmed the regression before the fix via a link failure for the missing export.

• git diff --check — passed.

• The full test suite and Dying Light 2 runtime were not tested.

Checklist

• Based on current upstream main; no duplicate PR implements this export/NID.
• One topic per PR; implementation and regression tests only.
• No code comments or silent stubs added.
• No notes, reports, images, or third-party code added.
• Generic behavior; not title-specific.
• Depends on: none.
• AI-assisted: yes (OpenCode).

[oneandonlydean]

Does Dying Light 2 (PPSA02262) stop at sceKernelFdatasync on main? Please add the log line where it stops; an import alone isn't enough for a new export. Also, sceKernel* functions return SCE error codes, so a bad descriptor should give 0x80020009, not -1 (the test at GuestFilesystem.cpp:158 should check that).

[github-actions]

Progress report (de5e541 → 57be03a)

➖ System libraries: 84.52% (+0.0%, +1 implemented functions)

✅ 1 implemented

Library	Name
libkernel	sceKernelFdatasync

[klosixgaming-arch]

Independent validation on Windows

Built and ran the PR head 8734b923 on my own box:

• Windows 11 Pro (26200), WinLibs MinGW-w64 GCC 15.2.0 (r7), Release, BUILD_TESTING=ON.

Results:

• guest_filesystem_tests PASSES (exit 0). libkernel.prx exports sceKernelFdatasync (verified with nm: T sceKernelFdatasync).
• The bundled GuestFont.cpp change: guest_font_tests PASSES (exit 0) after the font .prx DLLs are present in the tests dir.

Rigor check (is the test real?): I removed the sceKernelFdatasync implementation from Stdio.cpp, rebuilt, and confirmed guest_filesystem_tests fails to link — undefined reference to 'sceKernelFdatasync' — then restored it → PASS. So the new test genuinely requires the new export rather than passing unconditionally. This matches the "link failure for the missing export" base regression noted in the PR.

One concern — scope / "one topic per PR": this PR bundles two unrelated changes:

1. the sceKernelFdatasync feature + its GuestFilesystem.cpp test, and
2. an unrelated font-test isolation change in GuestFont.cpp (the PR's head commit is even titled "test(font): isolate system font failure case").

The author checked "One topic per PR", but these are two distinct topics. I'd suggest splitting the font-test change into its own PR so review/rollback stay clean.

Honest limits of what I verified:

• On Windows I verified the valid-fd == 0 path and the export. The invalid-fd == -1 assertion is inside #ifndef _WIN32, so it does not run on Windows — I could not verify that path here.
• I did not test the Linux/POSIX fdatasync path (no Linux here) or the Dying Light 2 title (no game/ROM).

Written with AI assistance.

[sktharunbalaji13-cmd]

Status update after the rebase and scope cleanup.

Title evidence (Dying Light 2, PPSA02262) — reported by @tv7 in #1794 (comment):

• Clean main: the loader stops at FAIL: unresolved ELF import 30Rh4ixbKy4 from libkernel.prx.
• With this change: that import resolves, and execution advances to FAIL: unresolved ELF import -q-vbO859Tw from libSceVideoOut.prx.

This is evidence the import blocker advances past 30Rh4ixbKy4; it is not a claim that the title becomes playable (12 further imports remain unresolved).

Scope — the PR now contains only core/libs/prx/libkernel/File/src/Stdio.cpp and core/libs/tests/GuestFilesystem.cpp. The bundled test(font) change was dropped, since main now carries that fix (SetFontDirectory in GuestFont.cpp), which had caused a redefinition build error.

Behavior — sceKernelFdatasync maps native failures through SceErrorFromErrno(errno) and returns 0 on success, matching the neighbouring sceKernel* file calls; the closed-descriptor assertion expects 0x80020009.

Tests — guest_filesystem passes on Windows (MinGW-w64 GCC 15.2.0) and on Linux.

[orishamir]

Built the PR (head 36bc9214) merged with current main and probed the new export against a guest socket: sceKernelFdatasync(socket) returns 0x80020009, while fdatasync_nid_postfix(socket) on the same descriptor returns -1 with errno 22, because the POSIX export checks GuestSockets::FirstDescriptor first and this one does not. The EBADF comes from the host rejecting 0x10000000 as an fd, so it is accidental; on Windows the same call reaches _commit and the UCRT invalid-parameter handler in CrashReport.cpp before returning.

Suggest moving the guard and the host call into sceKernelFdatasync and having fdatasync_nid_postfix wrap it, the way fsync_nid_postfix wraps sceKernelFsync.

Probe (linked like guest_filesystem_tests, Linux):

const int s = socket_nid_postfix(2, 1, 0);        // 0x10000000
sceKernelFdatasync(s);                             // -> 0x80020009
fdatasync_nid_postfix(s);                          // -> -1, guest errno 22
sceKernelFdatasync(0x7fffffff);                    // -> 0x80020009

Not run on Windows. AI-assisted.

[sktharunbalaji13-cmd]

Thanks @orishamir for catching the guest-socket descriptor issue — that was a real bug.

The fix is in 57be03a4:

• sceKernelFdatasync now checks GuestSockets::FirstDescriptor before any native syscall. An open guest socket returns SCE EINVAL (0x80020016); a closed or invalid guest socket returns SCE EBADF (0x80020009). This prevents high guest descriptors from reaching _commit on Windows (which would trigger the UCRT invalid-parameter handler).
• fdatasync_nid_postfix now delegates to sceKernelFdatasync through PosixResult, so the socket guard and error mapping live in one place — no duplicated logic.

The targeted guest_filesystem_tests passed on Windows/MinGW and Linux/WSL, including the new assertions for a valid socket (EINVAL), a closed socket (EBADF), and a high invalid descriptor like 0x7fffffff (EBADF). The full test suite and Dying Light 2 were not tested.

[orishamir]

Confirmed on 57be03a merged with current main (Linux): guest_filesystem_tests passes, and the probe from above now gives sceKernelFdatasync(socket) = 0x80020016 for an open guest socket and 0x80020009 for a bad descriptor, matching fdatasync_nid_postfix. Nothing further from me on this one.