Manage our own pushback state

[hadley]

Fixes #532. Closes #533.

[jcheng5]

I think there's a bug in the way I used pushBack to begin with, and persists in this PR. When blocking=FALSE, readLines will return partial lines, and AFAICT we can't distinguish between partial and complete lines. I just pushed a unit test for this.

[jcheng5]

Alright, I think this is ready

[hadley]

Should we apply the same logic to resp_stream_lines() so that you always get a complete line? Or maybe move the logic that there so that resp_stream_sse() can use resp_stream_lines()?

And does this mean we're drifting away from the idea of blocking vs non-blocking? Or maybe that distinction is no longer useful?

[jcheng5]

Should we apply the same logic to resp_stream_lines() so that you always get a complete line? Or maybe move the logic that there so that resp_stream_sse() can use resp_stream_lines()?

Uh... it looks like readLines already has the correct behavior--it only returns complete lines, even in nonblocking mode. I'll add a test for this.

Could we have used resp_stream_lines to create resp_stream_sse all along? Was all that logic in resp_stream_sse for nothing!? 😞

And does this mean we're drifting away from the idea of blocking vs non-blocking? Or maybe that distinction is no longer useful?

No, I don't think so? I think we still need blocking for convenience, and non-blocking for async to be possible.

[jcheng5]

Sorry, I was wrong about readLines not returning partial lines. So pushback would be good here.

[jcheng5]

Well, that turned out to be quite an adventure. It's good to go now, AFAICT.

[jcheng5]

Sorry that my tests added seconds of latency to the test suite. Is there a way to nudge webfaker from the main testthat process, so we could get rid of those sleep statements?