Closed recursion-ninja closed 4 years ago
@Boarders, take a peak at the hpack-generation
branch. If you think it's a step forward, merge it into master
.
I'll give it a look.
Also, builds with cabal
are failing because cassava
is messed up, and cassava
is a dependency of criterion
...
Builds with stack
work because of some dependency graph black magic.
I am not too keen on giving up cabal builds so we might want to hold off on a merge until it is solved.
I agree. However I don't think hpack
caused this issue, I think it just brought it to our attention again by requiring all dependencies for all build targets. I think that if you try and build any build target that depends on criterion
you will encounter this problem. We can test my hypothesis by trying to build the pcg-file-parsers:bench-time
build target on master
, which depends on criterion
and does not use hpack
.
Yes, I just don't think we should break the ability to build any library with cabal until it is fixed. Currently we mostly will be incapable of building benchmarks but not the inability to build the project itself nor individual libraries. I don't think it is worth just giving that up.
Yeah, again I agree. I think the real solution is probably having the library maintainer of cassava
fix their library instead of us trying to repeatedly apply bandaids. I can open an issue when I return.
I opened an issue: https://github.com/haskell-hvr/cassava/issues/177
I got information from @RyanScott to resolve the cassava for us.
We should explore #158, and decide between which approach offers the easiest maintainability.
Decided on #158 over this approach.
We repeat ourselves a lot in our various sub libraries. One such example is the GHC flags which are the same in every library.
hpack
promises to reduce duplication.See branch
hpack-generation
for changes.Invoke the following to generate the
.cabal
files:$ make run-hpack