Avoid using enable-overloading flag

[hamishmack]

Automatic flags are not automatic in stack. So we have to manually specify them for each gi-* package. I think it might be possible to work around this by eliminating the enable-overloading cabal flag.

Haskell gi-*.cabal files contain:

if flag(enable-overloading)
     cpp-options: -DENABLE_OVERLOADING
     build-depends: haskell-gi-overloading == 1.0.*
else
     build-depends: haskell-gi-overloading == 0.0

We could change this to just:

build-depends: haskell-gi-overloading < 1.1

and replace ENABLE_OVERLOADING with haskell-gi-overloading_MIN_VERSION(1,0)

Instead of -f-enable-overloading we would need to use --constraint=haskell-gi-overloading==0.0.

In stack.yaml just having

extra-deps:
- haskell-gi-overloading-0.0

Would be enough to switch off overloading.

Another option might be to keep the enable-overloading flag but only on one package (perhaps haskell-gi). So haskell-gi.cabal would have the automatic flag and:

if flag(enable-overloading)
     cpp-options: -DENABLE_OVERLOADING
     build-depends: haskell-gi-overloading == 1.0.*
else
     build-depends: haskell-gi-overloading == 0.0

While all the gi-*.cabal files would have:

build-depends: haskell-gi-overloading < 1.1

and use haskell-gi-overloading_MIN_VERSION(1,0). The stack.yaml would be:

extra-deps:
- haskell-gi-overloading-0.0

flags:
  haskell-gi:
    enable-overloading: false

This is not quite as nice.

CC @mgsloan

[mgsloan]

Would it be reasonable to have default: false in the flag definitions for overloading? I believe that this would fix your usecase too, but doesn't help much for the true case.

This would cut down boilerplate, though. Interesting way to share a "flag" among packages.

[garetxe]

This sounds reasonable enough. The whole machinery for disabling overloading is there mostly for the benefit of leksah currently, so whatever makes your life simpler is good with me.

If you have some time and want to send a pull request I would be happy to merge. Otherwise I will get to it when I get some time. It should be a very easy patch, but I am very pressed on time these days.

Would it be reasonable to have default: false in the flag definitions for overloading?

It would certainly be a valid choice. But I would prefer to keep the current default for overloading, since it leads to a nicer API (subjective, but I prefer to use the overloaded API). And this flag business is just a workaround for the slowness of GHC, hopefully at some point the root problem will get fixed.

[garetxe]

This is now done in 4fce037. Please take a look, and let me know if it works for you, or you see anything that does not look right.