You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Two tests report a build failure as a PASS (t.Skipf on 'cannot build cmd/rbgo'), plus the full env-var-gate inventory #687
If cmd/rbgo stops compiling, both report skip, the lane is green, and the
binary-level HTTP and socket coverage silently stops being exercised. t.Skipf
here says "this machine cannot build it"; what it actually reports is "the
product is broken". Those are opposite meanings and only one of them should be
green.
They should be t.Fatalf. There is no environment in which this repo's own cmd/rbgo legitimately fails to build — every CI lane builds it, and a developer
who cannot build it wants to know.
This is the same defect class as #682, found the same way
#682 was a guard that could not run (behind RBGO_BUILD_IT=1, which no workflow
set) and so failed open for three days. This is a guard that runs and reports
success on the failure it exists to catch. Both are "a check whose silence is
indistinguishable from a pass".
The sweep that found these also turned up the rest of the picture, worth
recording here so it is not rediscovered one at a time:
testing.Short() (3 sites): no workflow passes -short, so these do run.
The inverse case, listed for completeness.
Capability-probe skips (11): symlink, /dev/zero, mkfifo, user.Current(), Geteuid()==0. Satisfied on today's runners; they would go
quiet rather than red if that changed.
Platform skips (~24): all three OS lanes exist, so every branch is reached.
2 "gap recorded" skips in scripts/conformance/webapp/webapp_test.go:131,177
— that package is in ./..., so a removed binding presents as a skip.
1 skip that can never fire, internal/vm/grape_internal_test.go:171:
Activity
test: no test may skip on a failure to build, and a guard says so (#812)