Cabal omits true branch in conditional

[cstrahan]

Pandoc's cabal file gets mangled by jailbreak-cabal; this:

Executable make-pandoc-man-pages
  Main-Is:       make-pandoc-man-pages.hs
  Hs-Source-Dirs: man
  Build-Depends: pandoc,
                 base >= 4.2 && < 5,
                 directory >= 1 && < 1.3,
                 filepath >= 1.1 && < 1.4,
                 old-time >= 1.0 && < 1.2,
                 time >= 1.2 && < 1.6
  Default-Language: Haskell98
  if flag(make-pandoc-man-pages)
    Buildable:   True
  else
    Buildable:   False

is turned into this:

executable make-pandoc-man-pages

    if flag(make-pandoc-man-pages)
    else
        buildable: False
    main-is: make-pandoc-man-pages.hs
    build-depends:
        pandoc -any,
        base -any,
        directory -any,
        filepath -any,
        old-time -any,
        time -any
    default-language: Haskell98
    hs-source-dirs: man

Note: this is using jailbreak-cabal compiled with GHC 7.10.

(for reference: http://hackage.haskell.org/package/pandoc-1.13.2/revision/2.cabal)

[abbradar]

This also breaks cheapskate build, which is needed for Idris.

[abbradar]

haskell/cabal#2598

[cstrahan]

@abbradar Thank you for submitting the issue upstream.

[peti]

Interestingly enough, the problem doesn't seem to occur when building jailbreak-cabal with Cabal 1.18.x.

[abbradar]

One possible workaround for now can be using jailbreak-cabal from packages.ghc784; would it be hard/too hacky?

[peti]

It's possible. It would just trigger a lot of re-builds, I suppose, so it's kind-of expensive.

[abbradar]

Yeah, already noticed that. I have come with a rather hacky solution instead: abbradar/nixpkgs@c877e29 . Do we want this ugly thing mainlined or rather wait till the proper fix? OTOH this allows us to build affected packages.

[abbradar]

NM, I looked at how pandoc problem is avoided and think I'll just use sed too.

[peti]

I pushed peti/nixpkgs@daeefd2 to have test builds run at http://hydra.cryp.to/jobset/nixpkgs/haskell-updates.

[peti]

I've edited the jailbreak-cabal page on Hackage to constrain the Cabal version to 1.18.x: http://hackage.haskell.org/package/jailbreak-cabal-1.3/revisions/.

[peti]

Well, the version built with Cabal 1.18.x causes build failures, too: http://hydra.cryp.to/eval/457274#tabs-still-fail. The refcount package illustrates the issue: http://hydra.cryp.to/build/870257/log/raw. I'm not yet sure what's going on there, but those packages built correctly when we were using jailbreak-cabal with Cabal 1.22.x.

[abbradar]

Ouch! Does using Cabal 1.20.x fix this? If not, I can investigate! (fix: this should have been an exclamation)

Another fix: Tested on refcount; looks like it's good now.

[peti]

OK, building jailbreak-cabal with Cabal 1.20.x gives us the best performance for the time being. I'll merge the updates to master in a minute.

[peti]

Contrary to what commit haskell/cabal@e6bff69 says, this issue does not appear to be fixed in Cabal 1.22.4.0. When I build jailbreak-cabal with GHC 7.10.2, then it breaks the pandoc.cabal file just like old versions used to.