Set read buffer to 0

[cboden]

refs reactphp/event-loop#26

[cboden]

Courtesy of @mbonneau

[cboden]

I'd like to merge this by June 8th unless any valid objections. ping @clue @WyriHaximus

[clue]

All other methods test the Stream in isolation by mocking the environment (unit tests), while this test uses the environment (integration test). Perhaps move this to a separate test file? (Also, there's a missing newline)

[clue]

The code looks good to me and I'm all for getting this in! 👍 We're already handling our own buffers, so it makes sense to use unbuffered I/O on a lower level.

Given how long it took to trace this (subtle) bug, we should take the time and make sure to not introduce any regressions. In particular:

• How does this affect file I/O?
• How does this affect TCP server sockets (socket component)?
• How does this affect TCP clients (socket-client component)?
• How does this affect SSL/TLS clients (socket-client component)?
• Performance regressions?

[clue]

So far this only affects reading from the stream – how about also disabling the write buffer?

[cboden]

I'm leaning towards "if it ain't broke don't fix it". The read buffer is fixing a specific problem. I'm all for trying to set the write buffer to 0 in another PR if we can show it increasing performance or fixing a bug.

[clue]

if it ain't broke don't fix it

Sounds fair :)

[kingcrunch]

@cboden What's missing for a merge?

[WyriHaximus]

LGTM 👍

[mbonneau]

+1

[davidwdan]

👍

[clue]

Given how long it took to trace this (subtle) bug, we should take the time and make sure to not introduce any regressions […]

We have yet to spot any regressions, so I see no point in holding this off any longer. Let's get this in 👍

[clue]

Note that this PR introduced an incompatibility with HHVM which is being addressed in #28.