haskell_cabal_library: Add haddocks

[thufschmitt]

Build and export the haddocks for haskell_cabal_libraryes

This builds on top of #1097 because the cabal_wrapper part would have to
be rewritten otherwise (though it's minor and easy to port to master if
#1097 gets delayed)

[Profpatsch]

Rebased on master.

[thufschmitt]

@Profpatsch I've fixed the failure. Waiting for the CI to confirm, but it should be good now

[Profpatsch]

@regnat do you still want to merge this?

[thufschmitt]

Well that's up to you ;)

I just raised the fact that it makes the build of the cabal packages slower. But if nobody's stepping against it then yes, please merge

[mboes]

Haddock does add a fair amount of time to build times. Is there some way to make this optional?

[thufschmitt]

Haddock does add a fair amount of time to build times. Is there some way to make this optional?

I don't think we can't make it like haskell_library where haddocks are built on-demand (because all the build is happening in a single bazel action and I'm not sure we can reasonably split this as that would mean messing up with cabal's internal buisness). But we can add an extra argument to haskell_cabal_library to indicate whether or not we want the haddocks to be built − at the cost of a bit of complexity in the code as this has some repercussions both in the build script and in the exported providers

[mboes]

mmh, let's start by merging this. Provided we have an idea of the impact. Could you collect numbers about how much slower bazel build @stackage//stack is?

[thufschmitt]

So I benchmarked building @stackage//:proto-lens (because stack isn't exported in rules_haskell) both on this branch and on master, and strangely enough it's notably faster on this branch:

Branch	Mean [s]	Min [s]	Max [s]	Relative
haskell_cabal_library	135.147 ± 4.914	131.490	148.002	1.0
master	186.213 ± 3.220	183.164	193.916	1.00

I guess there has been another change on master making the build slower, I'll see whether I get the same results after a rebase

[thufschmitt]

(ftr the above results have been generated with hyperfine --export-markdown "benchs_on_branch_$(git rev-parse --abbrev-ref HEAD).md" -p "bazel clean --expunge && rm -rf ~/.cache/bazel_disk_cache" "bazel build @stackage//:proto-lens")

[thufschmitt]

I did some new benchmarks after rebasing on top of master, and indeed we now have a slight slowdown on this branch:

Branch	Mean [s]	Min [s]	Max [s]	Relative
haskell_cabal_library (rebased)	199.437 ± 3.021	196.493	206.374	1.00
master	186.213 ± 3.220	183.164	193.916	1.00

So this makes a 7% slowdown on this example, which I think is acceptable

(on a side note, I'd like to know why master seems to be much slower than what it was some times ago. Maybe that's because of #1156 )