[Monotone-devel] FYI: potential fix for log_quits_on_SIGPIPE failure

[Zack Weinberg]

For awhile the buildslaves have been getting mysterious failures on
log_quits_on_SIGPIPE.  I looked at it briefly last night and in more
depth this morning, and I *think* I have fixed it.  However, as the
problem was a race condition, I am not 100% sure I reproduced the
problem the buildslaves were having.  So I've gone ahead and pushed
the fix to mainline, and we'll find out :-)

I saw some 0.29-release prep go by on IRC, and I hope that committing
to mainline at this time has not screwed anything up.

(The curious are encouraged to read
log_quits_on_SIGPIPE/__driver__.lua and tell me if anything doesn't
make sense - the test will break if operations are not done in the
right order, and avoiding all races involves a second pipe that may
appear unnecessary on first glance, so it's important that the
commentary is clear to future maintainers.)

zw