[Based on a look at the code, would appear to be a real problem.
Fixes: (1) child reads from the pipe before exiting.  But this is a
race, presumably can't work; (2) child tells the parent "I'd like to
exit" and then parent tells child "OK, I've turned it off".  Might
work, haven't thought too hard; (3) Just catch the SIGPIPE in the
parent.  This seems simple but I don't really like it.  I guess maybe
because it wouldn't distinguish between SIGPIPE for this reason and
SIGPIPE for other reasons?  Or general gut feelings?

-kingdon]

Date: Fri, 18 Sep 1998 18:32:38 -0700
From: Xeno <xeno@goto.com>
To: bug-cvs@gnu.org
Subject: cvs server flow-control bug

There's a problem with the flow-control done by the cvs server process I
thought people might be interested to know about.  Workaround is easy,
just comment out the #define SERVER_FLOWCONTROL in src/options.h and put
a rebuilt executable on the server side.

The problem is that when the parent process tries to write a
flow-control signal to the child process, there's no guarantee that the
child process will run long enough to read it.  When updating a
particularly large file (29M) with cvs 1.9 on irix 6.2 server-side, it
looks like the parent process blocks on a flow-control write() and the
child process doesn't read() it before exiting.  So the parent process
gets a SIGPIPE from that write() and freaks, reporting an error back to
the cvs client.

Hope this helps somebody somewhere, thanks for creating such a cool
tool.

Xeno