tree 1fb626720c49cc1dea2433f687858b7f1e1590be
parent 1b27761f1cd4840a914b5255e4357385091562e5
author Brad Fitzpatrick <bradfitz@golang.org> 1446090554 +0000
committer Brad Fitzpatrick <bradfitz@golang.org> 1446138697 +0000

http2: fix Transport's flow control control when writing request bodies

Adapation of Blake's proposed fix in https://golang.org/cl/16463
with a few changes:

-- bug fix (advance the buffer after writing)
-- don't reacquire/release the buffer in the loop. it was done like
   that in case the max frame size changed while writing. Instead, push
   that down into awaitFlowControl since it has to acquire that lock
   anyway. Now it returns between 1 and the lower of how much we read
   and how much we're allowed to write.

This does mean that if we start a request with a max frame size of
32KB, we'll never write larger than 32KB frames until the the next
request (because our scratch buffer we read into is only 32KB), but we
will start writing smaller DATA frames immediately once we see the
peer's SETTINGS frame.

Change-Id: I47fc503062f9602fe448cf7a36fc500e5d6b8ef9
Reviewed-on: https://go-review.googlesource.com/16443
Reviewed-by: Blake Mizerany <blake.mizerany@gmail.com>