Internal libraries break `stack haddock`

[ryanreich]

The issue reported in haskell/cabal#4969 appears not to be fixed in stack: building haddock documentation for a package with internal libraries causes an error while resolving dependencies of the main library. The resolution proposed there, to use cabal new-haddock, has no obvious stack equivalent.

[istathar]

I've been following links to issues between projects and the long chain eventually led me to #3955 which is merged. But this issue is still occurring (as of lts-11.7 and stack-1.7.1) so I'm at a bit of a loss.

    Preprocessing library for haddock-library-1.4.5..
    Running Haddock on library for haddock-library-1.4.5..
    Cabal-simple_mPHDZzAJ_2.0.1.0_ghc-8.2.2: internal error when calculating
    transitive package dependencies.
    Debug info: []

[mihaimaruseac]

Thanks for the report, I was able to reproduce and I'll try to solve it (I think it's close to a solution for #3996)

[istathar]

@mihaimaruseac Trying to build pandoc, for what it's worth. I'm sure there's a simpler minimal reproducible case.

[mihaimaruseac]

There is, I reproduced it from the example in #3996 (although there is no documentation there to be generated)

[istathar]

Does anyone know of a workaround? Should revert to stack-1.6? I'm quite unable to build my projects right now because of this.

[mihaimaruseac]

Sorry, I stopped working on it to get HCAR out. I'll restart working on this after Tuesday/Wednesday when i hope to get HCAR public.

[istathar]

@mihaimaruseac certainly. Most appreciated. I'm just a bit surprised that I couldn't find an obvious way to get around this. Usually in stack if something goes wrong it's worst-case nuke the .stack-work directory. :). Look forward to trying any code you want me to test if/when the time comes.

[NorfairKing]

I just bumped into this problem as well. I haven't found a workaround.

[NorfairKing]

@mihaimaruseac, any luck?

[mihaimaruseac]

Not yet, still working on it.

[mihaimaruseac]

I made #4033 for #3966, next step is this. With the repo to reproduce #3966 I have a successful stack haddock if the internal library is a dependency of the main library but not if not.

[mihaimaruseac]

@afcowie, do you recall which version of Pandoc exercised this bug? I tried it now with the latest but it no longer has internal libraries so it works for me.

[mihaimaruseac]

I found a workaround: with the fix in #4033 and if the internal library is not a dependency of the main library, stack haddock finishes successfully.

I'll keep looking into this, but so far it seems that both in the case when the sublib is a dependency of the main library and not, Stack calls cabal haddock with the exact same arguments, after properly building the code. Surely I'm missing something.

[istathar]

do you recall which version of Pandoc exercised this bug?

The one in resolver lts-12.20. I just tried still happening six months later

$ stack haddock pandoc

with lts-12.20 and no change. Which resolver are you using, @mihaimaruseac?