It's impossible to build cassava with GHCHEAD post-#161

[RyanGlScott]

After #161, one cannot build the cassava library using a GHCHEAD configuration. Here is an example of a build failure from lens's GHC 8.6.1 Travis build matrix entry:

Failed to build cassava-0.5.1.0.
Build log (
/home/travis/.cabal/logs/ghc-8.6.0.20180714/cassava-0.5.1.0-7e18f6cd87ea4b7167810387f8520b18f7b969c5a859e28b2a3cf23ef7b6a170.log
):
Configuring library for cassava-0.5.1.0..
Preprocessing library for cassava-0.5.1.0..
Building library for cassava-0.5.1.0..

Data/Csv/Conversion.hs:26:3: error:
     error: #error **INVARIANT BROKEN** Detected invalid combination of `text-short` and `bytestring` versions. Please verify the `pre-bytestring-0.10-4` flag-logic in the .cabal file wasn't elided.
     # error **INVARIANT BROKEN** Detected invalid combination of `text-short` and `bytestring` versions. Please verify the `pre-bytestring-0.10-4` flag-logic in the .cabal file wasn't elided.
       ^
   |
26 | # error **INVARIANT BROKEN** Detected invalid combination of `text-short` and `bytestring` versions. Please verify the `pre-bytestring-0.10-4` flag-logic in the .cabal file wasn't elided.
   |   ^

Data/Csv/Conversion.hs:99:0: error:
     warning: "MIN_VERSION_text_short" is not defined [-Wundef]
     #if MIN_VERSION_text_short(0,1,0)
     ^
   |
99 | #if MIN_VERSION_text_short(0,1,0)
   | ^

Data/Csv/Conversion.hs:99:0: error:
     error: missing binary operator before token "("
   |
99 | #if MIN_VERSION_text_short(0,1,0)
   | ^

Data/Csv/Conversion.hs:1010:0: error:
     warning: "MIN_VERSION_text_short" is not defined [-Wundef]
     #if MIN_VERSION_text_short(0,1,0)
     ^
     |
1010 | #if MIN_VERSION_text_short(0,1,0)
     | ^

Data/Csv/Conversion.hs:1010:0: error:
     error: missing binary operator before token "("
     |
1010 | #if MIN_VERSION_text_short(0,1,0)
     | ^
`gcc' failed in phase `C pre-processor'. (Exit code: 1)
cabal: Failed to build cassava-0.5.1.0 (which is required by
criterion-1.5.0.0). See the build log above for details.

@phadej suspects this is because we are passing --allow-newer=bytestring.

[hvr]

It's a known issue with the heuristic of --allow-newer for the < and <= operators (which necessitates the use of --constraints on cabal flags to workaround); I was worried about such scenarios, hence why I added those CPP safeguards so we'd notice when this happens.

Anyway, a pragmatic short-term workaround is to add a tweaked .cabal file to head.hackage

[RyanGlScott]

Anyway, a pragmatic short-term workaround is to add a tweaked .cabal file to head.hackage

Sounds reasonable to me. Do you mind preparing such a patch? I'll admit that I don't feel confident enough to perform surgery on cassava.cabal myself, given the complexities involved here.

[hvr]

@RyanGlScott I've added something to the head.hackage index right now (not yet pushed to the head.hackage git repo); can you give it a try and let me know if it helps?

[RyanGlScott]

Thanks @hvr. With that patch, I am able to build cassava with a GHCHEAD configuration again (as demonstrated here).

[hvr]

I think we can close this one for now