Bug#1104335: britney/debci: sometimes lists a non-regression fail as a regression
Package: release.debian.org
The testing excuses list sometimes shows an autopkgtest as 'Regression',
with the reference link going to an old (often old enough that the log
is no longer available) passing migration-reference, when a newer
migration-reference exists and failed (i.e. when the failure is not
actually a regression).
In all the cases I've noticed, this happened on only one architecture,
even when the underlying 'reference used to pass, now fails' condition
exists on more than one architecture.
Instances I've noticed recently are r-bioc-keggrest/amd64 #1104334
blocking r-base, pique/(I've forgotten which architecture) blocking
r-base, and r-cran-plotly/riscv64 (listing
https://ci.debian.net/packages/r/r-cran-plotly/testing/riscv64/58034134/
as the reference) blocking r-cran-testthat. (The fact that these are
all R packages is probably because that's where I've been looking
recently, *not* a property of the bug itself. This is not an unblock
request, as none of those are otherwise ready.)
Workaround: Retrying the _reference_ test (using the link on the testing
excuses page) makes it stop counting as a regression. Waiting *may*
also do so.
Reply to: