Re: [Chicken-users] Calling out to C causing hang

[Chris Double]

On Mon, 07 Apr 2003 09:40, felix wrote:
> handle_interrupt is only called by C_raise_interrupt. C_raise_interrupt is
> either called by C_check_for_interrupt(in chicken.h) or by UNIX signal-
> handlers.

handle_interrupt is only called by C_reclaim I think. C_raise_interrupt sets a 
flag (interrupt_reason) that gets checked in C_reclaim to then process it. Is 
that right?

> Are you using finalizers? Or multithreading?

No to both.

> The best way of tackling this seems to be to walk through the whole
> thing with gdb, setting breakpoints in handle_interrupt and
> C_raise_interrupt (and C_reclaim). It would be nice if you could send me
> the gdb output starting from the point where handle_interrupt is called and
> up to the minor GC loop.

Unfortunately I haven't been using gdb. What's the easiest way of getting a 
debug build of chicken? Is there something I pass to configure or make to set 
the '-g' flag to gcc?

What I've been doing is putting lots of 'printf' debugging in runtime.c and 
doing a make of everything. It's slow to build though! When I run my test 
case I redirect the debug output to a file. It gets to about 58MB or so 
before I manage to kill it while it's looping.

I'm not sure what you what to see but if you let me know I can produce it. 
Currently it looks something like:


handle_interrupt called from C_reclaim (interrupt_reason 255)
c_save_and_reclaim before call to C_reclaim
C_reclaim, handle_interrupt not called, tgc is 360
C_fromspace_top: 1078298468
C_fromspace_limit: 1078333760
We are not jumping to full_gc
Before line of code labeled  /* Clear the mutated slot stack: */
Completed  /* Mark live values: */ loop
After heap_scan loop, gc_mode is GC_MINOR
Setting tgc to new value (milliseconds - tgc)
Performing longjmp to C_restart with value of 1.
c_save_and_reclaim before call to C_reclaim
C_reclaim, handle_interrupt not called, tgc is 360
C_fromspace_top: 1078298512
C_fromspace_limit: 1078333760
Not full gc
We are not jumping to full_gc
Before line of code labeled  /* Clear the mutated slot stack: */
Completed  /* Mark live values: */ loop
After heap_scan loop, gc_mode is GC_MINOR
Setting tgc to new value (milliseconds - tgc)
Performing longjmp to C_restart with value of 1.

[...The following loops for thousands of times with the exact same data...]
c_save_and_reclaim before call to C_reclaim
C_reclaim, handle_interrupt not called, tgc is 360
C_fromspace_top: 1078298512
C_fromspace_limit: 1078333760
Not full gc
We are not jumping to full_gc
Before line of code labeled  /* Clear the mutated slot stack: */
Completed  /* Mark live values: */ loop
After heap_scan loop, gc_mode is GC_MINOR
Setting tgc to new value (milliseconds - tgc)
Performing longjmp to C_restart with value of 1.

[...Then a slight change, the tgc is now 400, but again repeated thousands of 
times...]
c_save_and_reclaim before call to C_reclaim
C_reclaim, handle_interrupt not called, tgc is 400
C_fromspace_top: 1078298512
C_fromspace_limit: 1078333760
Not full gc
We are not jumping to full_gc
Before line of code labeled  /* Clear the mutated slot stack: */
Completed  /* Mark live values: */ loop
After heap_scan loop, gc_mode is GC_MINOR
Setting tgc to new value (milliseconds - tgc)
Performing longjmp to C_restart with value of 1.

And so on forever with only the tgc value changing. Any pointers to other 
places to put debug info? How can I find out where that longjump is going to? 
Eventually it gets to C_save_and_reclaim but I'd like to find out the trace 
for where that gets called. If I get chicken built with debug so gdb works on 
it I'll get a backtrace.

Chris.