SIGTRAP leaks when using backquotes and object allocation profiler

[doudou]

The whole interpreter aborts with SIGTRAP when using the object allocation profiler. Simply creating the following bla.rb file

`ls`

and running

CPUPROFILE_OBJECTS=1 CPUPROFILE=acd-profile-allocs ruby -rperftools -S bla.rb

triggers it on my machine (Debian unstable, ruby version)

ruby 1.9.3p194 (2012-04-20 revision 35410) [x86_64-linux]

[doudou]

Oh .. and the C backtrace is

#0  0x00007f6a4b13d8c1 in rb_newobj () at gc.c:1165
#1  0x00007f6a4b144666 in io_alloc (klass=16911960) at io.c:674
#2  0x00007f6a4b151056 in pipe_open (eargp=eargp@entry=0x7fff98094b40, prog=19933640, fmode=fmode@entry=1, convconfig=convconfig@entry=0x0, modestr=0x7f6a4b2507f6 "r")
    at io.c:5454
#3  0x00007f6a4b151680 in pipe_open_s (prog=19933640, fmode=fmode@entry=1, convconfig=convconfig@entry=0x0, modestr=0x7f6a4b2507f6 "r") at io.c:5522
#4  0x00007f6a4b151e4f in rb_f_backquote (obj=<optimized out>, str=19933640) at io.c:7514
#5  0x00007f6a4b21f443 in vm_call_cfunc (me=0x10a10e0, blockptr=0x0, recv=<optimized out>, num=1, reg_cfp=0x7f6a4a1c19e0, th=0xfc7520) at vm_insnhelper.c:404
#6  vm_call_method (th=th@entry=0xfc7520, cfp=cfp@entry=0x7f6a4a1c19e0, num=1, blockptr=0x0, flag=<optimized out>, id=<optimized out>, me=0x10a10e0, recv=16747000)
    at vm_insnhelper.c:534
#7  0x00007f6a4b2152d9 in vm_exec_core (th=th@entry=0xfc7520, initial=initial@entry=0) at insns.def:1015
#8  0x00007f6a4b21a964 in vm_exec (th=0xfc7520) at vm.c:1220
#9  0x00007f6a4b221e60 in rb_iseq_eval (iseqval=16741040) at vm.c:1447
#10 0x00007f6a4b12da60 in rb_load_internal (fname=16665000, wrap=<optimized out>) at load.c:310
#11 0x00007f6a4b12db69 in rb_f_load (argc=<optimized out>, argv=<optimized out>) at load.c:383
#12 0x00007f6a4b21f443 in vm_call_cfunc (me=0x10cb160, blockptr=0x0, recv=<optimized out>, num=1, reg_cfp=0x7f6a4a1c1bf0, th=0xfc7520) at vm_insnhelper.c:404
#13 vm_call_method (th=th@entry=0xfc7520, cfp=cfp@entry=0x7f6a4a1c1bf0, num=1, blockptr=0x0, flag=<optimized out>, id=<optimized out>, me=0x10cb160, recv=16963880)
    at vm_insnhelper.c:534
#14 0x00007f6a4b2152d9 in vm_exec_core (th=th@entry=0xfc7520, initial=initial@entry=0) at insns.def:1015
#15 0x00007f6a4b21a964 in vm_exec (th=0xfc7520) at vm.c:1220
#16 0x00007f6a4b221e60 in rb_iseq_eval (iseqval=28887960) at vm.c:1447
#17 0x00007f6a4b12da60 in rb_load_internal (fname=28874480, wrap=<optimized out>) at load.c:310
#18 0x00007f6a4b12db69 in rb_f_load (argc=<optimized out>, argv=<optimized out>) at load.c:383
#19 0x00007f6a4b21f443 in vm_call_cfunc (me=0x10cb0c0, blockptr=0x0, recv=<optimized out>, num=1, reg_cfp=0x7f6a4a1c1f08, th=0xfc7520) at vm_insnhelper.c:404
#20 vm_call_method (th=th@entry=0xfc7520, cfp=cfp@entry=0x7f6a4a1c1f08, num=1, blockptr=0x0, flag=<optimized out>, id=<optimized out>, me=0x10cb0c0, recv=16959200)
    at vm_insnhelper.c:534
#21 0x00007f6a4b2152d9 in vm_exec_core (th=th@entry=0xfc7520, initial=initial@entry=0) at insns.def:1015
#22 0x00007f6a4b21a964 in vm_exec (th=th@entry=0xfc7520) at vm.c:1220
#23 0x00007f6a4b221f4d in rb_iseq_eval_main (iseqval=iseqval@entry=18113280) at vm.c:1461
#24 0x00007f6a4b12aeca in ruby_exec_internal (n=0x1146300) at eval.c:204
#25 0x00007f6a4b12b96d in ruby_exec_node (n=n@entry=0x1146300) at eval.c:251
#26 0x00007f6a4b12d3fe in ruby_run_node (n=0x1146300) at eval.c:244

[tmm1]

Oof. I think its safe to say the allocation profiler is completely broken at this point. /cc #46