[JJenkx]

What

For a temporary stream containing a single byte, %2c%n in the bundled libc returns1, stores that byte, and writes1 to %n. The current secure stream scanner writes the byte but reports EOF and leaves %n untouched. Read narrow character fields directly from StreamSource so any nonempty short field counts as a conversion, suppressed fields advance the count without an assignment, and exact-width fields leave the following input available. Preserve capacity validation, UTF16 paths and all string paths.

Scope avoids the contiguous-string implementation in #9087, whose bounded string path already handles part of this case. No dependency on that PR; its stream fallback is unchanged. The ordinary host-backed fscanf/vfscanf short-field gap is still outside this change.

Hold the stream's recursive file lock throughout secure scanning so another reader cannot interleave between the character-field peek and consume operations. The lock is released on return and exception paths.

Tested

Linux x86_64 Release; final head ff25c7aba5f6a53215385cae3627b00556c4dd99, original base f0c8f26eec6edd346338e7c2d7f060ece52e8f50. All four focused suites (guest_formatting, guest_wide_formatting, guest_safe_errors, file_streams) execute and pass with zero skips. The final header-only placement adjustment leaves both the qualified production library and regression-test executable byte-identical.

The new concurrency regression runs both fscanf_s and vfscanf_s over a 1MiB character field. A second thread uses ftrylockfile and ftell to detect a lock becoming available before the field ends; it also checks lock release after return, complete output, %n and destination guards. It fails at the intended lock assertion with initial PR head 63af66d6c5e. With unchanged main it passes the locking check and then fails the existing short-field assertion. Both assertions pass with this head. This is a host/HLE regression check; bundled SDK multithreaded execution is not claimed.

The original two bundled libc modules agree on all 2,880 authored single-thread rows across sscanf/sscanf_s/fscanf/fscanf_s. Secure stream differences remain 165/720 →24/720, correcting 141 complete rows. After the locking correction, all2,880 outputs remain byte-for-byte identical to the previously qualified PR outputs. The other three API matrices are unchanged; the24 remaining secure stream rows concern literal-prefix exhaustion. No title speed, graphics, native game, GPU, Windows/macOS local or console validation claimed.

Conventions pass. Merge preview is clean against pinned current main f6fddd60d826831f70f7c62e25423610ee742bcc; Formatting.cpp, FileStreams.cpp, FileStream.hpp and RuntimeSupport.cpp are unchanged on main since the tested base. The1449-PR overlap scan found one include-placement collision with #13049; moving the includes resolves the current-main pairwise merge without changing either tested binary. The9 already-conflicted shared-file peers were manually reviewed as before.

Locking correction, exact pins, controls, full focused logs/JUnit and overlap resolution, SHA256 f1e7c08f61d6d91ce27a3c3a3334345750f0b1adf72b7f4b67311af00d11d39b. Original two-reference authored probe and synthetic rows, SHA256 0bef305e812549fd8408a9cb41123548c9f3ae5be70f73cb7b4c5eb60b0a421a. No SDK/game binaries or captured game data included.

Independent Windows validation reported by bunbeans2-arch for exact topic ff25c7aba5f6a53215385cae3627b00556c4dd99 merged into main f7f03daf52: Windows 11 Pro 22631, i9-9900K, MSYS2 UCRT64 GCC 16.1, Release. file_streams passes 5/5; keeping the new test while restoring Formatting.cpp to main fails 5/5 at the assertion that a short character scan retains the file lock through the field. Plain main with its own source/test passes 3/3. This is the contributor's run, not a new local Windows execution.

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; library compatibility only
• GPUs tested: n/a
• AI-assisted: yes

[MarcZX10]

Reviewed — looks good

The narrow stream %c path now treats a nonempty short field as a conversion, preserves suppressed-input counting, and leaves the next field available after exact-width reads.

What I ran

• Reviewed exact head 63af66d6c5e20c9d3755bb5d84e6970077a39724 against current main e7d8e353362935a1f00e893d901b1debc671f5ce; merge preview was clean at 99ffa586aeebd06a8c37d85c62859a47fe3586c8.
• Environment: Linux 6.1.0-54-amd64, x86-64, Python 3.11.2. No compiler was used.
• The author’s test baseline is f0c8f26eec6edd346338e7c2d7f060ece52e8f50; I checked that main has no changes to either touched file since that baseline.

git merge-tree --write-tree upstream/main refs/review/pr-13057-current
git diff --check upstream/main...refs/review/pr-13057-current
python3 tools/check_conventions.py --base upstream/main --head refs/review/pr-13057-current

Results

Check	Result
Merge preview	Clean; tree 99ffa586aeebd06a8c37d85c62859a47fe3586c8
git diff --check	Exit 0; no output
Conventions	Passed
Hosted CI	Eligibility, conventions, intake, and report pass; labels and platform jobs queued; Linux product job skipped
Author-reported focused tests	Four libc test targets pass 4/4 each on Linux; the author reports the new short-field assertion fails on unchanged main and passes with the patch. I did not run these build-dependent tests locally.

Command output

99ffa586aeebd06a8c37d85c62859a47fe3586c8
All conventions checks passed.

Negative control

Assertion	Change to code under test	Result
A one-byte stream read by %2c%n reports the converted byte and updates %n	Author tested unchanged main scanner logic	Author reports the new assertion failed on main; this is not an independently run result

Sources

• Short stream %c implementation
• Short stream field tests

Not verified

• I did not build or run FileStreams locally. Windows and macOS runtime behavior and console behavior were not independently tested in this session.

[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 7386	core/libs/prx/libc/src/Formatting.cpp core/libs/tests/FileStreams.cpp
PR 8150	core/libs/tests/FileStreams.cpp
PR 9340	core/libs/prx/libc/src/Formatting.cpp
PR 11831	core/libs/tests/FileStreams.cpp
PR 12057	core/libs/tests/FileStreams.cpp

[JJenkx]

@MarcZX10 Corrected a concurrency regression I found in the initial implementation: its separate peek/consume operations let another thread obtain the file lock midway through a character field. Unchanged main's host scan keeps that field locked. Final head ff25c7aba5f6a53215385cae3627b00556c4dd99 holds the recursive stream lock throughout secure scanning and releases it on exit.

The new observable try-lock/position regression fails on the initial 63af66d6c5e head. Against unchanged main, the locking part passes and the earlier short-field assertion fails. All four focused suites pass on the corrected implementation; all2,880 reference rows remain identical to the earlier qualification, with141 corrected rows and no field regressions. The updated description links the new negative controls, hashes and logs. No Windows/macOS/SDK-thread or native-title result is claimed.

This changes the approved source and needs re-review; it is not an optional test-only update.

AI-assisted: yes.

Hosted Windows follow-up on this exact head: the Windows job passes its changed host-only test-executable linking step. Its CTest execution step is skipped by workflow policy, so this supplies compile/link coverage and does not establish Windows runtime behavior. The four executed focused suites above remain Linux results.

[MarcZX10]

Re-checked — new head, looks good

The updated scanner holds the recursive stream lock across the whole field, addressing the interleaving window; the added regression checks lock availability during and after scanning, and current-main checks pass.

What I ran

Reviewed PR head ff25c7aba5f6a53215385cae3627b00556c4dd99, recorded PR base from the current PR metadata, against current main 91a69952c10099966378a6a9006b8ef616ae22da. The merge preview is clean (tree 342220ad34b27c990a1cf1529c4ee175a606b065). Environment: Linux x86_64, Git 2.39.5, Python 3.11.2; no compiler or project build was run.

git diff --check upstream/main...ff25c7aba5f6a53215385cae3627b00556c4dd99
python3 tools/check_conventions.py --base upstream/main --head ff25c7aba5f6a53215385cae3627b00556c4dd99
git merge-tree --write-tree upstream/main ff25c7aba5f6a53215385cae3627b00556c4dd99

Results

Check	Result
Current-main merge preview	Clean; tree 342220ad34b27c990a1cf1529c4ee175a606b065.
git diff --check	Passed; no output.
Conventions	Passed.
Focused Linux suites	Author reports all four suites pass on this head and 2,880 reference rows are unchanged; 141 rows are corrected with no field regressions.
New locking regression	The author reports the try-lock/position regression fails on the initial unlocked implementation; on unchanged main, locking passes while the earlier short-field assertion fails.
Hosted checks	Linux and Windows relinker jobs and Linux Debug/Release register-context checks passed; Windows passed. macOS relinker/register-context and write-watch jobs were still in progress at review time; product Linux job was skipped.

Unedited local command output

$ git diff --check upstream/main...ff25c7aba5f6a53215385cae3627b00556c4dd99
(no output; exit 0)

$ python3 tools/check_conventions.py --base upstream/main --head ff25c7aba5f6a53215385cae3627b00556c4dd99
All conventions checks passed.

$ git merge-tree --write-tree upstream/main ff25c7aba5f6a53215385cae3627b00556c4dd99
342220ad34b27c990a1cf1529c4ee175a606b065

Negative control

The author reports the new lock/position test fails on the initial implementation and that the prior short-field behavior remains failing on unchanged main; I did not independently mutate the code or rerun these controls.

Findings

Blocking

None found in the reviewed head.

Non-blocking

The Linux test results are author-reported. Windows hosted CI linked the changed test executable, but its CTest execution step was skipped; Windows runtime and native-title behavior remain unverified.

Sources

• Reviewed head
• Current-main merge base
• Author’s exact-head report and negative controls

Not verified

I did not build or run tests locally, run Windows/macOS CTest, or test SDK-thread/native-title behavior. Some macOS hosted jobs were still in progress at review time.

To get this to a good state

No further change is requested for this head. I will recheck the same commands and relevant tests if the PR head changes.

[MarcZX10]

Re-checked — new head, base SHA clarified

Correction to my review: the PR’s recorded base is e7d8e353362935a1f00e893d901b1debc671f5ce; I checked its merge preview against current main 91a69952c10099966378a6a9006b8ef616ae22da (tree 342220ad34b27c990a1cf1529c4ee175a606b065).

[bunbeans2-arch]

A Windows run with GCC 16: Windows 11 Pro 10.0.22631, i9-9900K, MSYS2 UCRT64 GCC 16.1.0, Release, this PR (ff25c7aba5) merged into main f7f03da.

• With this PR: file_streams passed 5 of 5.
• Same tree with only Formatting.cpp put back to main (this PR's test kept): it failed 5 of 5 at the new check Short stream character scan released its file lock during a field.
• main f7f03da alone (its own test and source): file_streams passed 3 of 3.

AI-assisted: yes (Claude Code)

[adriadam10]

Checked the %c stream path (short fields count as a conversion, suppressed fields advance the count, exact-width reads leave the rest available, capacity validation kept), the stream lock held through the scan and released on all exits, and the regression tests in file_streams; Conventions and Windows CI green; merged.