Hello all,
Unfortunately we discovered a major code generation regression
in GHC 9.12.3 that results in incorrect runtime results for
certain operations involving sub-word divisions.
We recommend avoiding upgrading to this minor release at this
time.
See https://gitlab.haskell.org/ghc/ghc/-/issues/26711 for more
details.
This is the same bug that affected GHC 9.12.1. A patch re-introducing
this bug was accidently backported onto the 9.12.3 branch
and went unnoticed throughout the testing period.
We are taking steps to ensure similar situations are less likely
to occur again, including looking at ways to improve our testing procedure
for minor GHC release, potentially using the Stackage infrastructure
to run the testsuites of a broad range of packages to detect
such issues.
We are also starting to prepare GHC 9.12.4 with a fix, which
we hope to release soon. However, a significant amount of time
has passed since the initial set of backports to GHC 9.12.3
were completed, which means that a number of merge requests
addressing issues of various severities has been marked for
backporting to the GHC 9.12 branch. We are in the process of
triaging this list and preparing a robust and reliable GHC 9.12.4
release.
Thanks,
Zubin.
