[MarcZX10]

What

Implements two of the reported wide-scanner differences in #9321, based on the console scanner behavior reported in merged #8479: an incomplete floating-point exponent such as 1e+ converts its mantissa, and a format literal at end-of-input returns the assignment count (zero when nothing was assigned) instead of EOF. Adds regressions for sized and unsized buffer scans and the stream scanner.

The reported short %c field behavior is not included: #9321 asks to confirm its return count, partial-output extent and precedence first. The libc debt entry records that remaining gap. This pull request does not close #9321.

Tested

Linux x86-64, Debian GCC 12.2, Release. Configured with cmake -S . -B build -G Ninja -DBUILD_TESTING=ON -DCMAKE_BUILD_TYPE=Release -DANYPS5_USE_SYSTEM_DEPS=OFF; built guest_wide_conversion_tests and guest_wide_stream_scan_tests; CTest passed guest_wide_conversion and guest_wide_stream_scan (2/2). No PS5 runtime or hardware test was performed.

python3 tools/check_conventions.py --base upstream/main reports zero errors and one debt-length warning on the existing libc scanner entry. The required local origin/main reference is 10,325 commits behind upstream; checking against it reports unrelated historical findings outside this change. tools/pr_overlap.py reports dependencies listed below.

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: fix(libc): end swscanf at %S and %C as a matching failure #8306, fix(libc): follow snprintf_s and sscanf_s constraints as measured on a PS5 #9087, test: organize library tests and label CTest suites #6044
• Fixes: none (the follow-up addresses two reported cases; libc: track reported wide scanf incomplete-field and literal-EOF differences #9321 remains open)
• Unblocks: none
• GPUs tested: n/a (CPU-only change)
• AI-assisted: yes

[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 3965	docs/dev/debt/libc.md
PR 5021	docs/dev/debt/libc.md
PR 6469	docs/dev/debt/libc.md
PR 6565	docs/dev/debt/libc.md
PR 6879	core/libs/prx/libc/src/ScanningWide.cpp docs/dev/debt/libc.md
PR 7081	docs/dev/debt/libc.md
PR 7447	docs/dev/debt/libc.md
PR 7648	docs/dev/debt/libc.md
PR 7715	docs/dev/debt/libc.md
PR 8150	docs/dev/debt/libc.md
PR 8965	docs/dev/debt/libc.md
PR 9001	docs/dev/debt/libc.md
PR 9038	docs/dev/debt/libc.md
PR 9133	docs/dev/debt/libc.md
PR 9170	docs/dev/debt/libc.md
PR 9408	docs/dev/debt/libc.md
PR 9447	docs/dev/debt/libc.md
PR 9526	docs/dev/debt/libc.md
PR 9527	docs/dev/debt/libc.md
PR 9669	docs/dev/debt/libc.md

[wachi35676]

Windows run, the platform the description doesn't list, plus a wider set of inputs compared with main and with FreeBSD 11.4.

Head c092916666 merged onto main a6dcea927 (clean). Windows 11 Home 10.0.26300, Ryzen 5 9600X, WinLibs GCC 15.2.0 (MinGW-w64 UCRT), Release, BUILD_TESTING=ON.

• guest_wide_conversion, guest_wide_stream_scan, guest_scan_count, guest_wide_formatting, guest_stdio_variadic, guest_wide_string: 6/6, 3 runs.
• Negative control: with ScanningWide.cpp from main, guest_wide_conversion fails and guest_wide_stream_scan aborts (0xc0000409).
• check_conventions.py: one warning, debt-length on docs/dev/debt/libc.md:128.

swscanf with %lf (destination -99), main → PR:

• 1e+, 1e, 1e-, 1e+x, 1ex, 0x1p, 0x1p+: 0 → 1, value 1. 1.5e: 0 → 1, value 1.5.
• -, ., e5, in, nan(, 0x: 0 on both. 1e+5, inf: 1 on both.

So the change is not limited to 1e+: every field ending in an exponent marker or sign now converts its mantissa, including 1e, 1ex and hex 0x1p. The updated debt entry says "other incomplete exponent forms are not separately verified", but their behaviour changes too. FreeBSD 11.4's parsefloat also commits to the last valid prefix, so it gives 1 for all of these.

Literals, main → PR:

• ""/x, ""/ x, ""/x%d, ""/%%: -1 → 0.
• x/x%d: -1 on both. /x: 0 on both. a/b: 0 on both. 5/%dx: 1 on both.

For a literal that meets end of input, FreeBSD 11.4's vfwscanf goes to input_failure and returns EOF when nothing was converted (vfwscanf.c L490-L495, L783-L784). C11 7.29.2.2 also calls running out of input an input failure. So the new 0 rests only on the firmware result reported in #8479; neither FreeBSD nor the standard supports it. The %% case changes too, so it would help to say whether #8479 covered %%.

AI-assisted: yes (Claude Code); results above were run on this machine.

[xXPBSXx]

Linux data point for c092916666 (WSL2, Ubuntu 24.04, kernel 6.18 - a virtualised kernel, not bare-metal Linux; GCC 13.3, Release, BUILD_TESTING=ON), this head merged onto main df2569d2c5 (clean merge). I built guest_wide_conversion_tests and guest_wide_stream_scan_tests and ran the executable directly (exit code, not through ctest):

• With the PR: exit code 0 for both.
• Without it (the PR's test sources on top of main, with core/libs/prx/libc/src/ScanningWide.cpp reset to main): guest_wide_conversion_tests exits 1 and guest_wide_stream_scan_tests aborts (SIGABRT, exit code 134).
  Not tested: macOS, Windows, the rest of the test suite.

AI-assisted (Claude Code).

[harisibrar1]

Checked against main 8b3e398 and FreeBSD 11.4 lib/libc/stdio/vfwscanf.c (release/11.4.0), plus the "Remaining differences" in #8479:

• The debt line seems to read main's "FreeBSD's vfwscanf reports an input failure for the first and converts the second" as being about 0x. On main, "the first" is the literal mismatch: vfwscanf.c L493-495 goes to input_failure when the character differs.
• For 0x with %x, FreeBSD's parseint pushes back the x and converts the 0 (L405-418, with PFXOK from L565). So 0x/%x returns 1 with 0 there, not an input failure.
• As written, the entry now says the wrong thing about 0x, and it no longer records that a/b%d returns 0 here but EOF on FreeBSD. Could the literal-mismatch divergence be kept, and the 0x clause say that FreeBSD converts it as 0?
• Small one: the literal-at-end-of-input sentence could also say it follows #8479 and that FreeBSD returns EOF there, as the exponent sentence does.
• #8306, #9087 and #6044 are all still open, and this PR merges and passes on main without them. If they are only shared-file overlaps, moving them out of "Depends on" might keep this from being held up.

Reviewed with AI assistance (Claude Code); the checks above were run locally on my machine.

[MarcZX10]

The new measurements in #9321 resolve the %% edge case too: the bundled libc.prx returned EOF for empty input with %%, while ordinary literals at end of input returned the assignment count. In this diff, Conversion() handles %% by calling the same Literal(u'%') method changed here to return Outcome::Matching on EOF. That means this change appears to make empty-input %% return 0, contrary to the newly reported firmware result.

Could the %% path preserve input-failure behavior while ordinary format literals use the measured assignment-count result, with buffer and stream regressions for both cases? The issue author identifies these as separate outcomes; the existing PR tests cover ordinary x but not %%.

[adriadam10]

Checked the incomplete-exponent and literal-at-EOF changes against the measurements in #9321 (bundled libc.prx, PPSA03671); tests fail without the change (also confirmed by @wachi35676 on Windows and @xXPBSXx on Linux).

Added on top:

• %% at end of input returns EOF again, as measured (Conversion() shared the changed Literal); tests for buffer, unsized and stream scans.
• Debt entry: the 0x clause now says FreeBSD converts the 0, and the literal-mismatch divergence from FreeBSD is kept (@harisibrar1's review).

merged.