[jotitadev]

What

guest_raw_resolver aborts on macOS with raw resolver: numeric canonical name differs: getaddrinfo with AI_CANONNAME|AI_NUMERICHOST on 127.0.0.1 or ::1 returns no canonical name.

FreeBSD 11.4's explore_numeric sets ai_canonname to the numeric host itself when AI_CANONNAME is asked ("based on a clarification in RFC3493"), and copy_ai carries it to every record. glibc sets it on the first record. macOS's getaddrinfo leaves it null for a numeric host, with or without AI_NUMERICHOST (a name such as localhost does get one). Measured with a plain C probe:

getaddrinfo with AI_CANONNAME	macOS 27.2	Linux (glibc)
127.0.0.1, ::1, datagram	null	127.0.0.1, ::1
127.0.0.1, any socket type	null, null	127.0.0.1, null, null

When the guest asks for AI_CANONNAME and the node is a numeric IPv4 (the same inet_aton parse gethostbyname uses) or IPv6 address, or AI_NUMERICHOST was given, a record the host left without a canonical name now gets the node's text. On macOS that fixes every record; on Linux it only fills the later records of a multi-record numeric answer, as FreeBSD does. A host-provided name is never replaced.

Tested

• macOS 27.2, Apple M2 Max, Apple clang, x86-64 Release under Rosetta 2, on main 801d5b9 plus the open macOS series (feat(libkernel): direct memory mappings on macOS hosts #7631, build(libkernel): compile Exception on macOS #11249, fix(libSceLibcInternal): declare environ for the spawn test on macOS #11358, build(libSceAudioIn): keep --exclude-libs off Apple's linker #10669, fix(libkernel): build the file, module and rtld calls on macOS hosts #10552, feat(libkernel): process, thread and clock queries on macOS hosts #7804, build(agc): compile Lifecycle, GuestMemory and GuestBufferMemory on macOS #11447, build(agc): keep the Linux-only parts of three AGC tests off macOS #11437, fix(agc): load the Vulkan loader and list MoltenVK on macOS #11554) needed to link.
  • Before: guest_raw_resolver aborts as above, 3 of 3.
  • After: it passes 3 of 3. 97 socket, poll, resolver and libSceNet tests give the same results with and without this change except guest_raw_resolver.
• Linux: Ubuntu 26.04, GCC 15.2, clean clone of this branch: guest_raw_resolver passes 3 of 3.
• tools/check_conventions.py passes.

Overlap: #5021 touches Resolver.cpp only in gethostbyname, which main already has; git merge-tree merges this branch with #5021 without a conflict in that file.

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 for the change. Building the test on macOS needs the open macOS series listed above.
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[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/libScePosixForWebKit/Resolver.cpp

[MarcZX10]

I found one edge case in the numeric-name detection. The fallback recognizes IPv6 only when inet_pton(AF_INET6, node, ...) succeeds, which excludes scoped literals such as fe80::1%en0; I verified inet_pton rejects the scoped form while getaddrinfo accepts it. FreeBSD 11.4 handles the %scope suffix separately and passes the original hostname as the canonical name to its numeric-address path (see explore_numeric_scope in getaddrinfo.c). Could you cover scoped IPv6 with AI_CANONNAME too, or narrow the stated behavior? I have not reproduced the missing canonical name on macOS for this scoped form, so this is a coverage concern rather than a confirmed target failure. The focused Linux/macOS reports and existing checks otherwise look coherent; no independent resolver test was run locally.

[jotitadev]

@MarcZX10 thanks, covered in b97c5b6 (new commit, no force-push).

• The numeric check now runs inet_pton(AF_INET6) on the part before % and keeps the whole node as the canonical name, as explore_numeric_scope passes the original hostname on.
• guest_raw_resolver gets a scoped case: fe80::1%1, raw hints, with AI_CANONNAME and with AI_CANONNAME|AI_NUMERICHOST, expecting fe80::1%1 as the canonical name.

What I measured with a C probe:

• macOS 27.2: getaddrinfo accepts fe80::1%1 and leaves ai_canonname null. Its inet_pton also accepts the scoped form, so the previous head already covered this case on macOS.
• Linux (glibc): inet_pton rejects the scoped form, as you saw, but glibc fills ai_canonname (fe80::1%1) itself.

So the change mainly stops the check from depending on how the host's inet_pton treats the suffix.

Results:

• macOS (x86-64 under Rosetta, main 801d5b9 + the macOS series): with only the scoped check moved first and without this PR, it fails with scoped numeric canonical name differs 3 of 3. At b97c5b6 guest_raw_resolver passes 3 of 3.
• Linux (Ubuntu 26.04, GCC 15.2, clean clone of b97c5b6): guest_raw_resolver passes 3 of 3.
• tools/check_conventions.py passes.

Written with AI assistance (Claude Code). I ran the builds and tests above on my machine.

[MarcZX10]

The scoped IPv6 follow-up addresses the reported case and adds the requested regression. I spotted one allocation-failure edge in the new fallback: after getaddrinfo succeeds, constructing std::string address can throw std::bad_alloc before freeaddrinfo(native) runs. This function has no surrounding catch/RAII cleanup, so the exception can cross the C ABI and leak the host addrinfo list. Please consider using a bounded stack buffer for the IPv6 portion or an RAII owner for native (and map allocation failure to the function’s normal error result). I did not reproduce OOM; this is a source-level exception-safety finding. The focused resolver test passed under QEMU TCG on the exact head, but this local host required emulation because its CPU lacks the project’s AVX2/FMA requirement.

[jotitadev]

@MarcZX10 thanks, done in c2a143da80 (new commit, no force-push). The address part before %scope is now copied into a 64-byte stack buffer and passed to inet_pton, so nothing between the successful host getaddrinfo and freeaddrinfo allocates or throws. A node whose address part does not fit is not a numeric IPv6 literal (the longest textual form is 45 characters), so it falls through to the existing path.

Linux (Ubuntu 26.04, GCC 15.2, Release): guest_raw_resolver and guest_resolver pass 3/3 on this head, including the scoped fe80::1%1 case from the previous commit, and git diff --check is clean. I did not rerun macOS for this commit. The change only replaces the std::string with the stack buffer and does not change the result: the same bytes reach inet_pton on both hosts.

Written with AI assistance (Claude Code). I ran the builds and tests above on my machine.

[maskym]

Windows result for head c2a143da80, merged into main f759e165d9 (the PR's Tested section names no Windows run).

Host: Windows 11 26300, Intel i7-12700K, WinLibs GCC 15.2.0 (x86_64-ucrt-posix-seh, r6), Release, BUILD_TESTING=ON, Ninja. Build clean.

The tree also carried the open PRs #11945, #11899, #11772, #11087, #10789, #11627, #11354, all merged cleanly, so the run is a joint one; the tests of this PR are guest_raw_resolver (0.04 s), all passed (the batch's 14 tests: 14/14). Not run on a PS5.

AI-assisted: yes (Claude Code), the runs and the numbers above are from this machine.

[MarcZX10]

Re-reviewed — approved

The stack-buffer follow-up resolves the allocation-failure path I flagged, and the scoped numeric IPv6 case is now covered.

What I checked

• Reviewed exact head c2a143da8050d1253f83abcf709072287919dd0b against live main 10bec5c322eb581b3e959e6c7e18bcf09e068e96.
• The new fallback strips %scope only for the inet_pton(AF_INET6) classification, keeps the original node as the canonical name, and uses a fixed 64-byte stack buffer. It introduces no allocation between successful getaddrinfo and freeaddrinfo.
• Host-provided canonical names retain precedence. The existing numeric IPv4 fallback and the AI_CANONNAME / AI_NUMERICHOST mappings are consistent with the new branch.
• Current-main merge preview is clean. The touched resolver and test files have not changed on main since the PR base.

Results

Check	Result
Current-main merge preview	Clean (92237d620cf093f8529753146a38b1b9cdf8c33e)
Diff check	Pass; no output
Conventions	Pass (All conventions checks passed.)
Focused Linux tests	Author reports guest_raw_resolver and guest_resolver passed 3/3 on the latest head
Focused Windows tests	Independent report says the PR test passed in a 14/14 test batch on the exact head, merged locally with its stated main and prerequisite PRs
Focused macOS tests	Author reports 3/3 on the earlier functional head; the final change only replaces the temporary string with a stack buffer and was not rerun on macOS
Hosted checks	Windows, conventions, labels, eligibility, and report pass; Linux and specialized product jobs skipped

Raw local output

merge-base: 801d5b9bc11049e6274daca09d70690bd5b8d0a9
merge-tree: 92237d620cf093f8529753146a38b1b9cdf8c33e
All conventions checks passed.
git diff --check: no output (pass)

Negative control

The author reports the original macOS failure 3/3 and the fixed resolver test passing 3/3. Linux passes on the latest head. No separate mutation or forced-failure run is reported.

Findings

Blocking

None found in the scoped resolver change.

Non-blocking

I did not independently build or run the resolver tests. The runtime results above are author and independent-review reports. The final stack-buffer-only commit was not rerun on macOS.

Sources

• Resolver implementation at reviewed head
• Raw resolver regression at reviewed head
• Live main

Not verified

I did not run platform builds, CTest, title behavior, or console behavior locally. The macOS result applies to the preceding functional head, not the final stack-buffer-only commit.

To get this to a good state

No further change is required for the resolver correctness and exception-safety findings reviewed here.

[imellag]

Confirmed on a second Mac with today's main. Apple M4 Pro, macOS 27.2, Apple clang 21.0.0, x86-64 Release under Rosetta 2. The tree is support/macos-boot-1030c (05bd71a66b, main 21a556f567 + the open macOS PRs), without and with head c2a143da80 merged (clean merge):

• Base: guest_raw_resolver aborts at numeric canonical name differs in 3 of 3 runs. It was 1 of the 19 failures in a full CPU-only ctest run on that tree; guest_resolver and call_site_resolver pass.
• This PR: guest_raw_resolver passes 5 of 5 (--repeat until-fail:5), and guest_resolver and call_site_resolver still pass.

AI-assisted: yes