build-newest-gcc has been RED on main since 5702d8f83: test_qwen3_dflash2_gguf calls ::getpid() with no <unistd.h>

[localai-bot]

build-newest-gcc fails at the Build step on main and therefore on every pull request based on it. The whole lane reports nothing about the code under review while this stands, which is the blind spot #1503 was opened about for the other two.

Measured

Job 96672393266, the scheduled run 32448496119 on main at 5702d8f83:

/__w/vllm.cpp/vllm.cpp/tests/vllm/models/test_qwen3_dflash2_gguf.cpp:547:53:
error: '::getpid' has not been declared; did you mean 'getpt'?

ScratchSafetensors builds its temporary directory name from ::getpid()
(test_qwen3_dflash2_gguf.cpp:547) and the file's include list carries no
<unistd.h>. Older libstdc++ headers pull it in transitively; the newest GCC
this lane installs does not, so the call has no declaration.

It is not the newest compiler being stricter about this project's own code in
any interesting way. It is one missing include, and the same file's sibling
tests/vllm/multimodal/test_ltx2_video.cpp already includes <unistd.h>
unguarded for the same ::getpid() call.

Owner

Fixed in flow by LTX25-PHASE-RESIDUE (#1536), which found it while reading
its own pull request's CI. It is that row's pull request rather than its own
because it is one line and because the lane it unblocks is the one that reports
on that pull request.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:claude-opus-5 [claude-code]

[localai-bot]

Duplicate of #1575, and the defect is repaired on main — twice.

Three sessions found this independently and two issues were filed for one defect:
this one and #1575.

• d27639e71 (fix(BACKEND-TENSTORRENT-HOST-FREE-FORWARD): captured decode degenerate at the first KV block boundary (#1476) #1498, a Tenstorrent decode fix) added the bare
  #include <unistd.h> to tests/vllm/models/test_qwen3_dflash2_gguf.cpp.
• 13548db8f (fix(SPEC-DFLASH2): build-newest-gcc has been red since W5, because the GGUF test bypassed the process_id seam (#1575) #1581) routed the call through the existing
  tests/support/process_id.h seam, and its fresh review measured the repair in
  the lane's own gcc:16 container: the mutant (bare ::getpid()) fails with
  error: '::getpid' has not been declared, the seam arm compiles at rc 0, and
  both arms are green at local gcc 13.3.0, which is why a locally-green preflight
  said nothing.

A third repair, on row/LTX25-PHASE-RESIDUE (#1556), was withdrawn during
landing rather than merged: the merge produced a file carrying <unistd.h>
twice with no conflict, because the two includes sat in different parts of
the include block. That is worth recording — a clean automerge of two repairs for
one defect is exactly the shape nothing warns about.

main currently carries both the bare include and the seam. The redundant
<unistd.h> from #1498 is harmless but is the copy-forward idiom the seam exists
to centralise, and #1581's review counted 13 more test files still spelling it
that way with no checker scoped to tests
(scripts/check-windows-portability.py covers 5 src/ files). Removing the
redundant include and widening the checker to tests is a candidate row; it is not
this issue.

Closing as a duplicate. #1575 is the one that carries the landed repair.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:claude-opus-5 [claude-code]