deadlock in 1.7.91 ?

deadlock in 1.7.91 ?

[Stanislav Ievlev]

Greetings!

I've tried to made guile module with glib's main_loop.

Following example below shows some strange interpreter freeze

--
# cat test.c
#include <glib.h>
#include <libguile.h>

SCM scm_test_mainloop()
{
        GMainLoop *loop= g_main_loop_new(NULL, FALSE);
        g_main_loop_run(loop);
}

void scm_init_test(void)
{
        scm_c_define_gsubr("test-mainloop", 0, 0, 0,(SCM(*)())scm_test_mainloop);
}

---
# gcc -fpic -shared `pkg-config glib-2.0 --cflags --libs` test.c -o
# libguile-test.so
# LD_LIBRARY_PATH=`pwd`  guile
guile> (+ 1 2)
3
guile> (dynamic-call "scm_init_test" (dynamic-link "libguile-test.so"))
guile> (+ 1 2)
3
guile> (define z (begin-thread (test-mainloop)))
guile> +
#<primitive-generic +>
guile> (+ 1 2)
....
<freezing here>

--

In strace log we see:
--
....... <guile eats (+ 1 2) here> .............
23004 read(0, "\r", 1)                  = 1
23004 write(17, "\n", 1)                = 1
23004 rt_sigprocmask(SIG_BLOCK, [INT], [RTMIN], 8) = 0
23004 ioctl(16, SNDCTL_TMR_STOP or TCSETSW, {B38400 opost isig icanon echo
...}) = 0
23004 rt_sigprocmask(SIG_SETMASK, [RTMIN], NULL, 8) = 0
23004 rt_sigaction(SIGINT, {0xb7e17d90, [], SA_RESTART}, {0xb7e17d90, [],
0}, 8) = 0
23004 rt_sigaction(SIGTERM, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGQUIT, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGALRM, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGTSTP, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGTTOU, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGTTIN, {SIG_DFL}, {0xb7e17d90, [], 0}, 8) = 0
23004 rt_sigaction(SIGWINCH, {SIG_DFL}, {0xb7e17d90, [], SA_RESTART}, 8) =
0
23004 close(16)                         = 0
23004 munmap(0xb7fd1000, 4096)          = 0
23004 close(17)                         = 0
23004 munmap(0xb7fd0000, 4096)          = 0
23004 time(NULL)                        = 1140441331
23004 rt_sigprocmask(SIG_SETMASK, NULL, [RTMIN], 8) = 0
23004 rt_sigsuspend([] <unfinished ...>
...... <main guile thread suspends and never returns> ...
23005 poll([{fd=10, events=POLLIN}], 1, 2000) = 0
23005 getppid()                         = 23004
23005 poll([{fd=10, events=POLLIN}], 1, 2000) = 0
23005 getppid()                         = 23004
23005 poll([{fd=10, events=POLLIN}], 1, 2000) = 0
23005 getppid()                         = 23004
23005 poll([{fd=10, events=POLLIN}], 1, 2000) = 0
 .... <etc.>
--

---
With best regards
Stanislav Ievlev.



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: deadlock in 1.7.91 ?

[Stanislav Ievlev]

Hi!

What about this problem?

--
Stanislav Ievlev.



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: deadlock in 1.7.91 ?

[Stanislav Ievlev]

On Sun, Feb 26, 2006 at 12:58:46PM +0300, Stanislav Ievlev wrote:
> Hi!
> 
> What about this problem?
Problem was in hidden select in g_main_loop (thanks to myself)

We really need enter_guile(), leave_guile() for such cases.
with-guile/without-guile will not work.

Are you planning to fix this problem?

--
Stanislav Ievlev.




_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: deadlock in 1.7.91 ?

[Ken Raeburn]

On Mar 6, 2006, at 4:27, Stanislav Ievlev wrote:
> On Sun, Feb 26, 2006 at 12:58:46PM +0300, Stanislav Ievlev wrote:
>> Hi!
>>
>> What about this problem?
> Problem was in hidden select in g_main_loop (thanks to myself)
>
> We really need enter_guile(), leave_guile() for such cases.
> with-guile/without-guile will not work.
>
> Are you planning to fix this problem?

But scm_leave_guile can result in garbage collection bugs in  
multithreaded programs.  Look in threads.c for a copy of the email  
message I wrote on the problem.  (There may be similar issues with  
scm_enter_guile, I haven't thought as hard about it.)  I doubt this  
can be fixed with that API, but if you spot something I missed, I'd  
be glad to hear about it.

The list archives seem to be missing some messages at the end of  
February.  Why will with/without-guile not work for you?  In the  
message dated 20 Feb in the archive, can you not call g_main_loop_run  
via scm_without_guile?

Ken


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel