michael.cadilhac@lrde.org (Michaël Cadilhac) writes:

> Stefan Monnier <monnier@iro.umontreal.ca> writes:
>
>> BTW, did anyone actually test my code to make sure that it does indeed work
>> and fix the problem?
>>
>>         Stefan
>
> As far as I tested, it doesn't work and freeze my Emacs. I'll
> investigate today.

I fixed what we discussed here.

I found two bugs in your proposal:
- the list was set to Qnil at the wrong place (wrong init)
- `return' was used in the handler, but it could still have some
   process to treat (a continue would have been better but still
   wrong).

A patch is attached.

However, the whole thing still doesn't work ; Emacs freezes or prints
this kind of error message: 

Error in post-command-hook: (wrong-type-argument listp (nil DEAD . 18764377))

I thought that the race condition could be the following:
- delq makes some change on  the list but not all,
- It is interrupted by the handler,
- The handler now has to deal with a malformed list.

It's certainly a bug too, as with the delq I had some Abort in the
handler of the form:

#0  0xb7a45951 in kill () from /lib/libc.so.6
#1  0x080eb75c in abort () at emacs.c:464
#2  0x081536b2 in Fsignal (error_symbol=137610425, data=137557813) at eval.c:1616
#3  0x081402ce in wrong_type_argument (predicate=137622913, value=149282605) at data.c:124
#4  0x0815c175 in Fmember (elt=71112, list=149282605) at fns.c:1468
#5  0x0818577c in sigchld_handler (signo=17) at process.c:6390

Which I think I don't have without, however, it is not the only error,
as with this line deleted, I still have the previous error and
problems. For example, I don't think it's normal for emacs to receive
a SIGPIPE

#0  0xb7ba62b8 in write () from /lib/libpthread.so.0
#1  0x00000002 in ?? ()
#2  0x0810561e in emacs_write (fildes=12, buf=0x8d26178 "!\n", nbyte=12) at sysdep.c:3370
#3  0x08187afb in send_process (proc=150054228, buf=0x8d26178 "!\n", len=2, object=149437363) at process.c:5489
#4  0x08188274 in Fprocess_send_string (process=150054228, string=149437363) at process.c:5643
#5  0x081534cb in Ffuncall (nargs=3, args=0xbfea69b4) at eval.c:2905

... Alas, most of the backtraces only show Emacs blocked in select, 
pthread things or wait_for_termination.

If any hacker can help :-)


Maybe it's  not the good solution  to keep traces of  these. The other
possibility (if we  don't consider the wait_for_termination solution),
is to store the PID of the synchronous process ; but I'm confused with
compatibilities with DOS and Mac :-/

If we have the PID of the synchronous process, we can explicitly test
it in sigchld_handler.