Defect
windows-msvc-cpu and windows-msvc-vulkan fail on every open pull request
with a warning-as-error that has nothing to do with the change under test:
src\vllm\entrypoints\openai\server_main.cpp(1315,55): error C2220: the following warning is treated as an error
src\vllm\entrypoints\openai\server_main.cpp(1315,55): warning C4456: declaration of 'loaded' hides previous local declaration
The two declarations are both on main today:
server_main.cpp:1025 — std::unique_ptr<vllm::entrypoints::LoadedEngine> loaded, the text engine;server_main.cpp:1315 — std::unique_ptr<vllm::multimodal::SpeechEngine> loaded, the speech engine, nested inside the first one's scope.
It is the ONLY warning in the job; nothing else in the build is unclean.
This is not #645
#645 is the M_PI portability
regression in the three LTX2 sources. Different file, different detector,
different failure. The windows-msvc-* pair being habitually red under #645's
name is what let this one sit unattributed — a second cause hiding behind a
known one, which is exactly why "known-red" needs a matched-arm check rather
than a name.
Evidence it is pre-existing rather than any one lane's
Checked against a matched arm on three unrelated pull requests, none of which
touches the speech surface: #956, #950 and #939 all fail with the identical
C4456: declaration of 'loaded' hides. Reproduced independently on #963.
Because windows-msvc-* are PR-only (skipped on push, per
#584), main carries no
baseline that would have caught it, so it presents as "a red the current PR
caused" to every author in turn.
Required outcome
Rename the inner declaration so it no longer shadows. No behaviour change, no
detector weakened, no warning suppressed.
Defect
windows-msvc-cpuandwindows-msvc-vulkanfail on every open pull requestwith a warning-as-error that has nothing to do with the change under test:
The two declarations are both on
maintoday:server_main.cpp:1025—std::unique_ptr<vllm::entrypoints::LoadedEngine> loaded, the text engine;server_main.cpp:1315—std::unique_ptr<vllm::multimodal::SpeechEngine> loaded, the speech engine, nested inside the first one's scope.It is the ONLY warning in the job; nothing else in the build is unclean.
This is not #645
#645 is the
M_PIportabilityregression in the three LTX2 sources. Different file, different detector,
different failure. The
windows-msvc-*pair being habitually red under #645'sname is what let this one sit unattributed — a second cause hiding behind a
known one, which is exactly why "known-red" needs a matched-arm check rather
than a name.
Evidence it is pre-existing rather than any one lane's
Checked against a matched arm on three unrelated pull requests, none of which
touches the speech surface: #956, #950 and #939 all fail with the identical
C4456: declaration of 'loaded' hides. Reproduced independently on #963.Because
windows-msvc-*are PR-only (skipped on push, per#584),
maincarries nobaseline that would have caught it, so it presents as "a red the current PR
caused" to every author in turn.
Required outcome
Rename the inner declaration so it no longer shadows. No behaviour change, no
detector weakened, no warning suppressed.