Replace blaze-builder with bsb-http-chunked

[sjakobi]

I have extracted Blaze.ByteString.Builder.HTTP into a new package.

[sjakobi]

This is the main change.

[sjakobi]

The Travis failure seems to be due to threads requiring base < 4.11.

[mightybyte]

Yeah, I already opened this PR. basvandijk/threads#15

@gregorycollins Thoughts?

[mightybyte]

@sjakobi Hmmm, when testing with GHC 8.4.1 cabal new-build --allow-newer works on your branch but cabal new-test --allow-newer fails. It looks like this is because the test suite depends on http-streams which depends on http-common which depends on blaze-builder. So we need to fix more than just the threads package.

[sjakobi]

Oof!

For http-streams, I have already opened aesiniath/http-streams#108. I'll wait for any feedback there until (maybe) I do more work.

http-common looks easy to fix but as with http-streams it's probably better to use a faster strict bytestring builder. I have opened aesiniath/http-common#14

[mightybyte]

Thanks for doing this work BTW. I was recently thinking that we should probably make the switch from blaze-builder to bytestring. Any chance you can modify bsb-http-chunked to not depend on bytestring-builder any more? It seems like it's mainly a shim package.

[sjakobi]

Any chance you can modify bsb-http-chunked to not depend on bytestring-builder any more? It seems like it's mainly a shim package.

Can you explain why that's important to you?

I use bytestring-builder only to stay compatible with GHC < 7.8. I can hide the dependency again for recent GHCs like this if you prefer.

[mightybyte]

Well, in general I like to minimize dependencies. I was thinking it wasn't buying us enough, but if it's buying us compatibility with < 7.8, then I think you're right, we should keep it for awhile longer. With snap we have a strong commitment to backwards compatibility. I just ditched 7.4 compatibility for snap-server rather reluctantly, so yeah...let's keep 7.6 compatibility for awhile.

[sjakobi]

I just ditched 7.4 compatibility for snap-server

Aren't you going to adjust your lower bound on base then?

[mightybyte]

Oh, good point. I had removed the Travis test, but not bumped the lower bound. Just did that.

[mightybyte]

@sjakobi It looks like your merge has an error in the cabal file.

(The master branch is now building with GHC 8.4 btw.)

[sjakobi]

I shouldn't have used the GitHub interface to fix the merge conflict… :)

BTW, now that blaze-builder has a new release, I leave it up to you whether you still want to do the switch. I do intend to modernize the internals and maybe speed up chunkedTransferEncoding a bit but that hasn't happened yet.