Avoid bytestring >= 0.11 with GHC < 8.0

[vdukhovni]

In bytestring 0.11 the PS constructor becomes a pattern synonym, only
available with GHC >= 8.0. Since this package depends on bytestring
internals, and in particular the PS constructor, with GHC < 8.0 it must
stick to older bytestring releases.

With GHC >= 8.0, this package will be compatible with the imminent
bytestring 0.11, so the dependency ceiling is for now set to < 0.12.

[vdukhovni]

Pushed an update for Cabal < 2.2, older Cabal versions don't support elif.

[fosskers]

I'm going to get better CI going with this repo, and then we can run this PR through it.

[chessai]

MASTER has been updated to use haskell-ci, a more automatic and consistent way of handling travis ci for haskell projects. more and more core libraries are moving toward using haskell-ci, and i think haskell-streaming should too.

[chessai]

@fosskers to answer your question about github ci, i would consider using it if there were an automated/stable way of setting it up. don't know of one (haven't googled and about to run out the door).

additionally, the core libraries convering on using haskell-ci + travis makes it particularly attractive to me. let me know if you have any thoughts.

[chessai]

Most recent master now builds with the new CI: https://travis-ci.org/github/haskell-streaming/streaming-bytestring

[fosskers]

I happen to have written about it: https://www.fosskers.ca/en/blog/github-ci

What I like about it is that it is fairly minimal, and has first-class support from the Github UI.

[fosskers]

@vdukhovni can you rebase this onto master so that CI kicks in again?

[vdukhovni]

@vdukhovni can you rebase this onto master so that CI kicks in again?

Done, I rebased both this PR and #18.

[fosskers]

Alright, CI is now testing a huge range of GHCs (including the ones relevant to this PR), and things look good, so let's get this in.

Thanks for this!

[vdukhovni]

Alright, CI is now testing a huge range of GHCs (including the ones relevant to this PR), and things look good, so let's get this in.

Thanks for this!

You're welcome. I should perhaps mention that in late-breaking news GHC 9.0 will require at least bytestring 0.11 and nothing older will build. Should that be reflected in the cabal files of packages using bytestring? For example, here we could also add a constraint on impl(ghc >= 9.0) requiring bytestring >= 0.11, or do we expect that to be handled automatically, or by other means?

http

Cc: @Bodigrim, @chessai , @bgamari