[Bug-readline] readline() in 7.0 seems to interfere with blocked SIGALRM

[Frédéric Brière]

[ I don't have much experience with either readline or signals, so ]
[ please forgive any ignorance (or stupidity) on my part.  :)      ]


readline 7.0 is causing a bug[*] in Twinkle, where we are normally
blocking SIGALRM in all threads, with one thread dedicated to calling
sigwait().  Upon calling readline(), SIGALRM now apparently becomes
unblocked (or caught and rethrown, I guess), resulting in the program
aborting.  This was not the case with 6.3.

This has been reported and confirmed on Debian/Ubuntu (amd64) so far.


I'm attaching a small example that reproduces this behavior:

 $ gcc -o readline_sigalrm readline_sigalrm.c -lreadline -lpthread 
 $ ./readline_sigalrm
 > Alarm clock

Note that disabling rl_catch_signals doesn't have any effect on this.
(This probably wouldn't be a good solution for us anyway.)

(I also noticed that this only seemed to occur when readline() was
called from the main thread.  I suppose this could very well be
OS-specific, though.)


Please let me know if there's anything I can do to help.  Thanks!

 [*] https://github.com/LubosD/twinkle/issues/89

readline_sigalrm.c
Description: Text Data