[bug#71427] Request for merging "haskell-team" branch

[bug#71427] Request for merging "haskell-team" branch

[Lars-Dominik Braun]

Hi,

I would like to merge the haskell-team branch. It contains a big quality
of life improvement for Haskell developers, by allowing cabal-install
to work with Guix-provided packages. Apart from that GHC was upgraded
to version 9.2.8 and Stackage was upgraded to the corresponding version
20.26.

Lars

[bug#71427] Request for merging "haskell-team" branch

[Liliana Marie Prikler]

Hi Lars,

Am Samstag, dem 08.06.2024 um 08:36 +0200 schrieb Lars-Dominik Braun:
> Hi,
> 
> I would like to merge the haskell-team branch. It contains a big
> quality of life improvement for Haskell developers, by allowing
> cabal-install to work with Guix-provided packages. Apart from that
> GHC was upgraded to version 9.2.8
So far LGTM.

>  and Stackage was upgraded to the corresponding version 20.26.
That commit really lacks a readable message, though.

Further, I think our work is far from done; just attempting to update
pandoc has me yak-shaving the depths of haskell-*.scm with no end in
sight.  If that's our only haskell application, we're in luck, but I
fear there might be more.

Cheers

[bug#71427] Request for merging "haskell-team" branch

[Lars-Dominik Braun]

Hi,

> >  and Stackage was upgraded to the corresponding version 20.26.
> That commit really lacks a readable message, though.

I know it does not conform to Guix standards. What do you suggest? One
commit per package is not feasible.

> Further, I think our work is far from done; just attempting to update
> pandoc has me yak-shaving the depths of haskell-*.scm with no end in
> sight.  If that's our only haskell application, we're in luck, but I
> fear there might be more.

Is Haskell-world, you can’t just update pandoc. Pandoc is part of the
curated set of Stackage packages and those *must not* be updated,
unless there is a new Stackage release which matches our GHC compiler
(there is not). Otherwise, well, dependency hell…

I’m also pushing a fix for ghc-aeson, which should fix most other
packages as well.

Lars

[bug#71427] Request for merging "haskell-team" branch

[Liliana Marie Prikler]

Am Mittwoch, dem 12.06.2024 um 21:53 +0200 schrieb Lars-Dominik Braun:
> Hi,
> 
> > >  and Stackage was upgraded to the corresponding version 20.26.
> > That commit really lacks a readable message, though.
> 
> I know it does not conform to Guix standards. What do you suggest?
> One commit per package is not feasible.
I think mentioning each file and package would be good enough™.
If parts can be split off into more commits, that's fine, but
everything that needs to be bumped along with the variable, needs to be
bumped.

> > Further, I think our work is far from done; just attempting to
> > update pandoc has me yak-shaving the depths of haskell-*.scm with
> > no end in sight.  If that's our only haskell application, we're in
> > luck, but I fear there might be more.
> 
> Is Haskell-world, you can’t just update pandoc. Pandoc is part of the
> curated set of Stackage packages and those *must not* be updated,
> unless there is a new Stackage release which matches our GHC compiler
> (there is not). Otherwise, well, dependency hell…
Is there any other way to get pandoc-3.2 then?  I have a package that
appears to depend on that.
Perhaps we could do -next variants, but there'd still be a lot of them.

> I’m also pushing a fix for ghc-aeson, which should fix most other
> packages as well.
Neat.

Cheers

[bug#71427] Request for merging "haskell-team" branch

[Lars-Dominik Braun]

Hi,

> I think mentioning each file and package would be good enough™.
> If parts can be split off into more commits, that's fine, but
> everything that needs to be bumped along with the variable, needs to be
> bumped.

I mean, if a 250-line-ish long, auto-generated commit message is
okay…?[1] It’s usefulness is disputable, but I can’t do this by
hand for ~230 packages.

> Is there any other way to get pandoc-3.2 then?  I have a package that
> appears to depend on that.
> Perhaps we could do -next variants, but there'd still be a lot of them.

You could try a fresh import with `guix import hackage -r pandoc`, but
you’ll probably run into issues with packages requiring more recent
GHC-provided packages. Then you can either try to build a newer GHC[2]
or download the entire Hackage database and use a solver to find a
combination of packages that works with our current GHC (i.e. what
Stackage does by hand, afaik).

> > I’m also pushing a fix for ghc-aeson, which should fix most other
> > packages as well.
> Neat.

This is done now. Let’s convince everyone else that skipping the merge
queue is fine in this case…

Lars

[1] git show | sed -nre 's#^\+\+\+ b/(.*)#* \1#gp; s/^@@.*\(define-public (.*)/(\1): Update./gp' | uniq
[2] https://issues.guix.gnu.org/issue/67921

[bug#71427] Request for merging "haskell-team" branch

[Liliana Marie Prikler]

Am Freitag, dem 14.06.2024 um 08:24 +0200 schrieb Lars-Dominik Braun:
> Hi,
> 
> > I think mentioning each file and package would be good enough™.
> > If parts can be split off into more commits, that's fine, but
> > everything that needs to be bumped along with the variable, needs
> > to be bumped.
> 
> I mean, if a 250-line-ish long, auto-generated commit message is
> okay…?[1] It’s usefulness is disputable, but I can’t do this by
> hand for ~230 packages.
Well, yes, we do have a script to do that, and it'd be an improvement
over what we already have.  Maybe review the output, but that'

> > Is there any other way to get pandoc-3.2 then?  I have a package
> > that appears to depend on that.
> > Perhaps we could do -next variants, but there'd still be a lot of
> > them.
> 
> You could try a fresh import with `guix import hackage -r pandoc`,
> but you’ll probably run into issues with packages requiring more
> recent GHC-provided packages. Then you can either try to build a
> newer GHC[2] or download the entire Hackage database and use a solver
> to find a combination of packages that works with our current GHC
> (i.e. what Stackage does by hand, afaik).
I've started doing that, but I get a conflict with a core package,
meaning that I'd need GHC 9.4 or newer.  We probably ought to look into
making that the current GHC to get a fresher Stackage.

> > > I’m also pushing a fix for ghc-aeson, which should fix most other
> > > packages as well.
> > Neat.
> 
> This is done now. Let’s convince everyone else that skipping the
> merge queue is fine in this case…
Note that I have little experience with Haskell otherwise, so whatever
I say, it'll probably not be too convincing.

Cheers

bug#71427: Request for merging "haskell-team" branch

[Lars-Dominik Braun]

Hi,

the branch has been merged into master.

Lars