From: Mathieu Bouchard <matju@...>
Date: 2003-04-19T14:17:29+09:00
Subject: Re: problem with rb_rescue2() ?


On Sat, 19 Apr 2003, Yukihiro Matsumoto wrote:

> |Now I want to ask you how much the ruby internals are resistant to a
> |non-VALUE being passed in a VALUE parameter? Especially the GC and such...
> |Is that looking for trouble ? (segfault ?)
> It's OK to pass non Ruby object pointer casting to VALUE.  Ruby's
> conservative GC can handle them.
> 							matz.

Good.

But in the meanwhile I thought about it and figured out I can use Fixnums
anyway -- my pointers are 0 mod 4 so i can >>2 them and then pass them to
INT2NUM() like I'm doing alot in other parts of my program.

But I have to report a problem I have. I'll try to make a summary of the
stack:

(eval):1: [BUG] rb_gc_mark(): unknown data type 0x20(0xbfffa560) non
object
ruby 1.8.0 (2003-03-31) [i586-linux]
Abandon (core dumped)

(gdb) bt
#0  0x40352a51 in kill () from /lib/libc.so.6
[...]
#5  0x404776fa in rb_gc_mark (ptr=3221202272) at gc.c:513
[...]
#13 0x4047834b in rb_gc () at gc.c:1240
#14 0x404769a7 in rb_newobj () at gc.c:323
[...]
#19 0x4049da49 in ruby_yylex () at parse.y:3295
[...]
#25 0x4045c00e in rb_eval_string (
    str=0x400142a0 "[\"ruby #{$!.class}: #{$!}\",*($!.backtrace)]")
    at eval.c:1382
#26 0x400137f3 in make_error_message ()
   from /home/projects/scite-ruby/scite-ruby.so
#27 0x400138a0 in rb_rescue_funcall_3 ()
   from /home/projects/scite-ruby/scite-ruby.so
#28 0x40464167 in rb_rescue2 (
    b_proc=0x40013804 <rb_rescue_funcall_2(unsigned long)>,
data1=3758085209, 
    r_proc=0x40013864 <rb_rescue_funcall_3(unsigned long)>,
data2=3758085209)
    at eval.c:4356
[...]

and then the rest are SciTE/Scintilla/GTK functions.

I'm also embedding Ruby in other apps and I'm having the same kind of
problems, though since it has the "heisenbug" type of behaviour, this bug
tends to be latent for long periods of time, be sensitive to small changes
to irrelevant portions of code, and tends to disappear while i try to
observe it. (e.g. the Valgrind debugger prevents the bug from happening at
all!)

My perfect nightmare.

________________________________________________________________
Mathieu Bouchard                       http://artengine.ca/matju