[Chicken-users] Re: reentrant signal handler? (was sigpipe problem)

[Lui Fungsin]

So I confirmed that the sigpipe hanging my chicken program stems from
the same problem - that the signal handler is not re entrant.

If I handle sigpipe, linux will send the signal when the pipe is
broken. Moreover, subsequent write to the same file no. will generate
another sigpipe.

If the signal sent are within a very short interval, it will hang the
chicken program (as demonstrated in my program below, which handle
sigint)


I compile the test program, then run it.

% ps -C test1
  PID TTY          TIME CMD
17767 pts/0    00:02:09 test1

Now if I send a signal to it, I can see that it does receive that by
typing "show" in the example program.

% kill -s INT 17767

Now, if I send multiple signint one after another one rapidly,

% kill -s INT 17767; kill -s INT 17767; kill -s INT 17767; kill -s INT 17767

This will send the chicken program into an infinite loop

% top

top - 13:44:47 up 11 days, 20:43,  2 users,  load average: 0.65, 0.22, 0.07
Tasks:  55 total,   3 running,  52 sleeping,   0 stopped,   0 zombie
Cpu(s): 98.0% us,  2.0% sy,  0.0% ni,  0.0% id,  0.0% wa,  0.0% hi,  0.0% si
Mem:    776088k total,   567476k used,   208612k free,    77948k buffers
Swap:  2096440k total,        0k used,  2096440k free,   370976k cached

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
17767 fungsin   25   0  4916 1832 1272 R 98.5  0.2   0:55.53 test1

And it looks like chicken is still doing some gc work while subsequent
signal arrives.

If my guess is correct, should the gc (or any non-entrant code) be
protected by a mutex or something?

TIA.
-- fungsin

% gdb test1 17767

0xb7e9da16 in C_reclaim () from /usr/local/lib/libchicken.so
(gdb) bt
#0  0xb7e9da16 in C_reclaim () from /usr/local/lib/libchicken.so
#1  0xb7e9e5ed in C_save_and_reclaim () from /usr/local/lib/libchicken.so
#2  0xb7e4dfef in f_2787 () from /usr/local/lib/libchicken.so
#3  0x0000001e in ?? ()
#4  0x00000005 in ?? ()
#5  0xb7ce3e5c in ?? ()
#6  0xbfdacdf4 in ?? ()
#7  0xbfdace04 in ?? ()
#8  0x0000001e in ?? ()
#9  0xb7eea934 in ?? () from /usr/local/lib/libchicken.so
#10 0xb7cd10f8 in ?? ()
#11 0xb7ce3e2c in ?? ()
#12 0xbfdacdf4 in ?? ()
#13 0xb7eea934 in ?? () from /usr/local/lib/libchicken.so
#14 0xb7ce3e5c in ?? ()
#15 0x00000005 in ?? ()
#16 0xb7cdfe88 in ?? ()
#17 0xb7e3a522 in tr4 () from /usr/local/lib/libchicken.so
#18 0x0804b2e0 in ?? ()
#19 0xb7eea934 in ?? () from /usr/local/lib/libchicken.so
#20 0x00a3fca0 in ?? () from /lib/ld-linux.so.2
#21 0x08049db8 in f_73 ()
#22 0xbfdace50 in ?? ()
#23 0xb7e9f353 in CHICKEN_run () from /usr/local/lib/libchicken.so
Previous frame inner to this frame (corrupt stack?)


On 5/1/08, Lui Fungsin <address@hidden> wrote:
> Hi,
>
> I know that the chicken core is not re entrant.
> In that context, what's the usual practice for writing posix signal
> handler in chicken?
>
> See example program below.
>
> If you type something, it will simply echo it.
> You can type 'show' to print the a global flag.
> Ctrl-C will set the flag to true.
> If you hit ctrl-c rapidly, you'll send it into a loop consuming 100%
> cpu. (ctrl-\ to quit it)
>
> Am I correct that the signal handler written in scheme will be invoked
> asynchronously?
> How would one write this program in chicken correctly? Maybe there's a
> way to do an atomic operation to set the flag?
>
> Thanks,
> fungsin.
>
> (declare
>  (uses extras regex posix utils srfi-13))
>  (require 'regex)
>  (require 'posix)
>  (require 'utils)
> (require-extension loop)
>
> (define *user-interrupted?* #f)
>
> (set-signal-handler!
>  signal/int
>  (lambda (signo)
>   (set! *user-interrupted?* #t)))
>
> (loop for l = (read-line)
>      while (and l (not (eof-object? l)))
>      if (string= l "show") do
>      (print *user-interrupted?*)
>      else if (string= l "exit") do
>      (exit 0)
>      else do
>      (write-line l))
>