When someone runs "bin/trial --reactor=select
childConnectionLost" on cygwin, using the cygwin version of Python,
either manually (which I've done on two cygwin installations), or from
the twisted buildbot --
http://buildbot.twistedmatrix.com/builders/cygwin-py2.5-select -- then
once the child process closes its fd #2, it subsequently receives no
further bytes from its fd #1, even though the parent process has written
some bytes to that pipe.
I can reproduce this by running the Twisted unit tests, like this:
step 1: get twisted:
svn co svn://svn.twistedmatrix.com/svn/Twisted/trunk twisted
step 2: cd into the top-level directory, and run trial:
cd twisted && python ./bin/trial --reactor=select
On other platforms that cygwin (see the Twisted buildbot --
http://buildbot.twistedmatrix.com/supported?branch=trunk -- then this
test exits quickly with success. On cygwin, using cygwin python.exe,
this test instead enter an infinite loop as the child waits to receive
further bytes on its stdin, and the parent waits for the child to
respond to those bytes. (This doesn't happen on cygwin with the standard
Win32 build of python.exe, of course, anymore than it happens on a
Windows system without cygwin installed when using that executable.)
I've opened a ticket about this on the twisted issue tracker as well:
http://twistedmatrix.com/trac/ticket/3529 # closing stdout in a child
process on cygwin means that process doesn't receive bytes from stdin
anymore. I think.
I've opened a ticket about this on the python issue tracker as well:
http://bugs.python.org/issue4295 # closing stdout in a child process on cygwin
means that process doesn't receive bytes from stdin anymore. I think.
The version of the cygwin package of python is 2.5.1-2.
Created attachment 3060 [details]
the output from cygcheck -s -v -r
Corinna Vinschen of cygwin requests a smaller test case:
Here is a smaller test case which shows incorrect behavior:
I will also attempt to attach demo.4.py to this ticket. There are more test
cases along with my notes about the behavior that they exhibited at:
Oh, I'm sorry, demo.4.py shows correct behavior. The difference between
demo.4.py -- which behaves correctly -- and "python ./bin/trial --reactor=select
childConnectionLost" -- which behaves incorrectly -- is not understood.