Closed bgamari closed 4 years ago
I've just released haskeline-0.8.0.0 off of the latest HEAD commit (ff19768affd99397468c5af4bfd81fd573088429).
Unfortunately, I then realized that 98d6924 doesn't correspond directly to a commit in judah/haskeline and its history is more than a year old. I pulled it from the GHC gitlab and compared the two:
https://gist.github.com/judah/2a09a824c2ed68bfdb2330f152594d00
The remaining changes appear harmless (one minor bug fix, a MonadFail instance, and some new functions that GHC doesn't care about). Is it too late in the release process to switch GHC to haskeline's HEAD? Otherwise I might need to re-release new major versions so we have one for GHC and one for the latest HEAD.
@judah, apologies for the belated response; this required a bit more investigation than I anticipated. Unfortunately, bumping to 0.8.0.0 directly won't be possible as the fact that the Cabal file defines both a library and executable component breaks the build system (specifically I've tested Hadrian, although I suspect the old make
build system is similarly afflicted). This is admittedly a deficiency of the build system although I'm afraid it's not one that can be fixed for 8.10.
There are two ways forward at this point (both, unfortunately, requiring a new release):
building-ghc
) which when enabled marks the executable component as buildable: False
Given that there appears to be some good bug-fixes in 0.8, I suspect the latter option is preferable.
Thanks @bgamari. The latter option makes sense to me. I'll aim to make a new Haskeline release either later today or tomorrow at the latest.
@bgamari I've pushed 463fc49d17bfab846cceba48bccc02ef285e6cba to the haskeline-0.8 branch. Please let me know whether that works for you; if so, I'll cut another release.
My local Hadrian build look good but the CI run is currently underway at https://gitlab.haskell.org/ghc/ghc/merge_requests/2404.
I saw that the validate step failed; at first glance it seemed like it wasn't passing the Cabal flag in the "validate" step.
Maybe it would be fine for you to just patch Haskeline locally in GHC and remove the problematic build/test components? I'd be fine with having that very slight difference in a release of GHC, since it wouldn't change the external API from the released version of haskeline-0.8.0.0.
The downstream GHC issue causing the validation failure has been resolved.
GHC 8.10.1 will ship with
haskeline
based on 98d69248d08389f349e12981da43797b8860ae67. It would be great to have a release based upon this commit sometime in the next week.