[atharvjairath]

What

On Linux installations with only C/C.UTF-8 locales, the alternate-locale URI escape/parser assertions added in #8017 could silently exercise no effective case and still report success. This follows up on the contributor's reproduction.

Register those existing assertions as guest_http_uri_locale. It reports the tested locale and explicitly skips with code 77 when no locale classifies Latin-1 0xE9 as alphabetic. Windows requires .1252. The ordinary HTTP test retains its C-locale assertions; production sources and every existing assertion expression remain unchanged.

CMake now prepares a private en_US.ISO8859-1 in the HTTP build directory on Linux using host localedef and installed locale sources. It runs from the checked-out PR tree, so a queue dispatch using the workflow from main executes it before registering the test. This addresses review 5687342073: the previous workflow-only setup never ran in full Build 40733382859, and the Linux skip audit correctly rejected the missing test coverage.

ANYPS5_HTTP_TEST_LOCALE_DIR remains an explicit override. Only guest_http_uri_locale receives LOCPATH. Missing localedef, unavailable source data or generation failure is reported during configuration; the test still probes available locales and skips with 77 if none is effective. No package downloads, installations or system-locale changes occur. The ineffective workflow step and its CMake argument are removed; build.yml now matches current main. check_skips.py is unchanged and still rejects lost coverage. BUILD.md documents the prerequisites and override.

The current overlap snapshot reports six base-conflicted peers. #9834 adds distinct transport/linkage/lifecycle work in HTTP CMake and GuestHttp; #674, #1835 and #5021 document input/native ARM64 work; #10194 and #11351 document GPU memory accounting and shader probe capture. Complete shared-file patches were retained and reviewed, using identical earlier scoped baselines for unchanged heads. Preserve all test registrations, the original Windows PATH block and documentation sections during integration. No locale duplicate, unmerged import or open prerequisite was found.

Tested

Exact committed head 384d3436deddce0c9995eeeebd1c47a292653af6, a fast-forward follow-up on 3eae29619679ce7fa3ef4f49f4f681fe396b02f7, with original base 8ff4baf53489033dae79090640143e82d159ee4d. The focused Release harness uses complete real libc and libSceHttp built at a41c and reused after reconfiguring this exact commit. Production and runtime fixture sources are byte-identical; both incremental builds report no work.

• Linux GCC 13.3 / native ARM QEMU 11.0.3 executing x86-64: the default configure path, with no LOCPATH, I18NPATH, workflow locale variable or explicit CMake locale directory, creates the private locale and passes 4/4 HTTP tests (0.58 s) through actual tools/select_tests.py ctest. The test reports alphabetic 0xE9; both the branch skip gate and current-main gate accept the JUnit with zero skips. The source-data fixture temporarily exposes the cached distro locales package's source files at /usr/share/i18n, then restores that absent path; no package is installed or system locale generated. A separate run sourcing those same files through I18NPATH passes 4/4 (0.55 s).
• Missing-source and missing-command controls both configure successfully, attach no private LOCPATH, and retain test skip 77. Both branch and current-main skip gates exit 1. The missing-command control runs after successful generation and proves stale generated files do not mask unavailability. An explicit locale-directory override works even with localedef absent; repeat automatic configuration succeeds. CTest JSON verifies exactly four HTTP tests and LOCPATH on only the locale test. locale -a remains C/C.utf8/POSIX.
• Windows PE, WinLibs GCC 15.2 UCRT / Wine 9 / QEMU 11.0.3: after exact-commit reconfiguration, 4/4 pass (13.42 s), including mandatory .1252. The external CTest manifest preserves arguments/skip metadata. No write-watch switch or runtime workaround is added.
• After-commit branch/current-main conventions, whitespace, full public-header audit (368 libraries / 12811 exports) and exact overlap CLI plus current-main-tool/shared-file review pass. The current-main-tool overlap snapshot uses main f21ef7c99c5af648c5a41bddb18ce118b8e60c46. Local merge previews pass on later main 82dfc54c7bba408caecaaf3ebc277b614779ef8a and preserve every owned patch payload; these previews are not later-main runtime validation. The initial find_program form failed the system-dependency convention; it was replaced with direct host-command invocation and the final commit was revalidated.
• The focused harness enables HTTP tests and disables libc tests. Its known whole-harness registration-audit limitation (nine pre-existing libc executables without registrations) is unchanged; this follow-up verifies the four HTTP registrations directly and does not claim a root full-project audit. Historical a41c mutation controls were not rerun because production and fixture sources are unchanged: the unchanged Linux fixture rejected locale-dependent ctype mutants, and the separate Windows diagnostic mutant reported high-byte URI rejection. Those retained external controls are not part of this PR.

For old head 3eae, the maintainer-dispatched full Build above merged into 7fea87746bde8e14a1d3341af85377c5b53f89ad: Linux built and ran CTest but failed the skip audit; the Windows job's build, runtime CTest and skip-audit steps succeeded. This is historical CI evidence for that head/merge. The new head needs a fresh maintainer dispatch. No local native Windows, full-project build, packaged runtime, title or PS5 validation is claimed. No current-head approval is inferred.

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

[oneandonlydean]

Thanks, the test reads right. One CI problem: the new guest_http_uri_locale returns skip code 77 when no Latin-1 locale is installed, and the Linux job's tools/check_skips.py (build.yml around line 547) fails on any skip that doesn't match its device-limitation patterns. Nothing in .github or tools provisions that locale, so once merged this would turn the Linux job red. (The PR's own CI skips the full Linux run, so the checks here can't show it.)

Could you either provision a private en_US.ISO8859-1 for the test (e.g. built at configure time and pointed to with LOCPATH), or add this skip message to the allowed patterns in check_skips.py with a short reason?

[MarcZX10]

The focused locale test design and its local evidence are detailed, but there is a blocking CI integration gap. On the Linux environment described in the PR (C/C.UTF-8/POSIX, with no locale that treats 0xE9 as alphabetic), guest_http_uri_locale returns 77 and skips. The PR itself notes that the CI skip audit rejects this unless the runner provides a Latin-1 locale, and the workflow does not currently provision one.

I checked the current main version of tools/check_skips.py as well as the version at the PR base. The current checker still rejects this skip: its device-limitation patterns do not recognize the emitted SKIP: URI locale regression... message. A representative CTest JUnit result with SKIP_RETURN_CODE=77 and that diagnostic returns error: guest_http_uri_locale skipped from the checker. The PR's automatic checks skip the full Linux build, so they do not cover this integration.

Please make the Linux test environment provide an effective private Latin-1 locale (as the PR's own validation does), or update the CI skip policy deliberately and document why this coverage skip is acceptable. Then run the relevant Linux CI path. I do not recommend approving while the full Linux job can fail on this test.

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 9834	core/libs/prx/libSceHttp/CMakeLists.txt L22

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 674	docs/dev/BUILD.md
PR 1835	docs/dev/BUILD.md
PR 5021	docs/dev/BUILD.md

Dismissing a stale review: the requested changes were addressed after it (new head) and the merge queue blocks on it. Please request changes again if anything is still open.

[adriadam10]

The full Build of this head (run 40733382859) fails in the linux job at python3 tools/check_skips.py build/ctest.xml: error: guest_http_uri_locale skipped.

• The job's step list has no "Prepare a private Latin-1 locale for the HTTP regression" step. The queue dispatches build.yml from main with pr=N, so the workflow edit in .github/workflows/build.yml (line 522) never runs for this PR; ANYPS5_HTTP_TEST_LOCALE_DIR stays empty and the test skips with 77, which the skip audit rejects.
• This means the provisioning can't be validated before merge, and the PR can't go green through the queue as written.

Suggested direction: do the provisioning from the PR's tree instead of the workflow, e.g. a script under tools/ that core/libs/prx/libSceHttp/CMakeLists.txt runs at configure time on Linux (generate en_US.ISO8859-1 into the build dir with localedef when the sources are installed, and set LOCPATH for guest_http_uri_locale only). Keep the skip-77 path for machines without localedef, and keep check_skips.py unchanged. The test and CMake registration changes themselves read fine.

[THEMANnt]

Reviewed 384d3436deddce0c9995eeeebd1c47a292653af6 against 8ff4baf53489033dae79090640143e82d159ee4d. The URI locale regression is separately registered with skip code 77. The Windows run actually selects .1252 and reports that 0xE9 is alphabetic, so its escape/parse checks exercise the intended locale condition. Source review covers Linux private-locale creation/LOCPATH handling; that Linux localedef branch was not executed locally. No actionable finding in this scoped test review.

Windows/GCC 15.3 Release: configured and built the actual pinned target; registered focused CTest: 2 passed, 0 skipped. Baseline was also built/run: 1 passed, 0 skipped. Matching cached dependency source pins and the matching existing FFmpeg package are used. No game, physical-console or GPU validation is claimed.

[adriadam10]

Checked the new head against the previous request: the locale is now generated during CMake configure from the PR tree (private en_US.ISO8859-1 via localedef, LOCPATH set only on guest_http_uri_locale), so it also runs when the queue dispatches the workflow from main. The test registration, skip code 77, mandatory .1252 on Windows and the PATH entry are in place, and the assertions are only moved; merged.