FS2 - Remove "elide empty chunks"

[diesalbla]

The FS2 library, by construction, guarantees that the uncons does not give, in the result, a non-empty chunk.

[rossabaker]

When did this become true? FS2-3 or FS2-2? Is this something that could also be done on an earlier release?

[diesalbla]

When did this become true? FS2-3 or FS2-2? Is this something that could also be done on an earlier release?

I found about this invariant this week, but it would be interesting to know how far it extends.

If one would like to find out, here is what has to be done: checkout the release of fs2 being used, look for the Output class. This is the only subclass of Pull that contains a chunk, and it is on this Pull constructor that compilation stops when using the .uncons function. If that Output class is private, and the only call to its constructor includes the guard not to create it with empty chunk, as in this line, then the uncons should not be giving out an empty chunk.

[rossabaker]

It appears that's true in fs2-2.5, in which case we could do the same all the way back to 0.21.

[rossabaker]

This looks good after the comment is stripped. I'll probably try to cherry-pick it to 0.21, unless you want to reopen with the same logic against that.

[rossabaker]

It cherry-picked cleanly to series/0.21! 🎉

[rossabaker]

Well, I should have run the tests instead of relying on reason and the compiler:

[error]   ! drain the epilogue with chunk size 3 (3 ms)
[error]    java.lang.RuntimeException: Chunk.empty.apply(0) (Chunk.scala:514)
[error] fs2.Chunk$EmptyChunk.apply(Chunk.scala:514)
[error] fs2.Chunk$EmptyChunk.apply(Chunk.scala:512)
[error] org.http4s.multipart.MultipartParser$.checkIfLast$1(MultipartParser.scala:312)

I'm going to revert the cherry-pick and merge the original.