REPL locks up on OS X 10.7

[fonnesbeck]

I get intermittent, unrecoverable REPL lockup when routinely using Julia on my OS X 10.7.3 machine (2011 MacBook Air). Most recently, just playing with various matrix calculations. Julia returned the value, but then accepted no further input. I was able to exit with ctrl-D.

[Keno]

Since you were able to quit with C-d, readline must be getting back control. It is probably an issue of it not being displayed properly, while the actual processing of the input is still being done.

[fonnesbeck]

I have had a recurrence of this problem just now, where I was unable to quit with ctrl-d. It appears to happen most reliably when switching to the terminal running Julia from another application. I am usually able to type a few characters, then it locks up. It happens a few times a day.

[JeffBezanson]

I think I can reproduce a variation of this on linux by hitting ctrl-z, then fg. From that point it doesn't work. Only happens with the readline frontend. See if you get the same problem with julia-release-basic.

[fonnesbeck]

Here's what happens when I restore a Julia process using fg on OS X:

Will try the basic release version.

[Keno]

@JeffBezanson Readline needs to be reinitialized after signals have been received (since the default readline signal handlers are disabled in repl-readline.c) - http://www.delorie.com/gnu/docs/readline/rlman_43.html

[JeffBezanson]

I think I have a fix in for the suspend and resume problem. Can people try it?

[pao]

It does work for the suspend/resume case, though it picks up exactly where it left off, so if you suspend in mid-expression and resume some time later you might have junk still hanging around you can't see. A more complete solution might be to reprint the julia> prompt and any text entered up to the point of suspend when it is resumed.

[fonnesbeck]

Works for me, thanks.

[Keno]

a call to rl_redisplay after rl_reset_after_signal is probably the best here (though only if were not in the middle of a computation). Looks like @JeffBezanson may have to bring the callback_en flag back ;)

P.S. sorry for not testing any of this. I'm currently testing something else that I can't interrupt by recompiling

[JeffBezanson]

Ok, done. Readline is just full of magic isn't it :)