_ProactorBaseWritePipeTransport.write() pauses the protocol if needed

_ProactorBaseWritePipeTransport.write() pauses the protocol if needed

[yselivanov, 2014-08-25 15:44:49]

Is it possible at all to have a unittest, which will show that your change is
actually needed?

[haypo_gmail, 2014-08-25 17:28:15]

On 2014/08/25 15:44:49, yselivanov wrote:
> Is it possible at all to have a unittest, which will show that your change is
> actually needed?

Hum, it's difficult to mock the write() method of each transport implementation,
because there are multiple transport implementations and each is very different.

Using real socket is not reliable, because all data may be written directly
depending on the speed of the computer, even if a local server is used and the
server doesn't read all written bytes.

The CPython test suite uses a limit of 16 MB + 1 byte to ensure that a send
blocks (Lib/test/support/__init__.py):
---
# A constant likely larger than the underlying OS socket buffer size, to make
# writes blocking.
# The socket buffer sizes can usually be tuned system-wide (e.g. through sysctl
# on Linux), or on a per-socket basis (SO_SNDBUF/SO_RCVBUF). See issue #18643
# for a discussion of this number).
SOCK_MAX_SIZE = 16 * 1024 * 1024 + 1
---

Here is an attempt to write such test:
http://codereview.appspot.com/126600043

The test pass on Linux, but fails on Windows because write() always completes
immediatly (even if the test writes 16 MB at once). I'm not sure that it's
possible to write a reliable and portable test this patch.