octplot 0.3.9 segfaults

[Nobody/Anonymous]

Folks,
  Thanks for making octplot!
On debian sarge, but with a home-built octave 2.71, I built octplot with gcc 4.0.4. As soon as I try to run line_demo, it segfaults.
  If I run under gdb and do a bt, I get:

Program received signal SIGSEGV, Segmentation fault.
0x411722d6 in free () from /lib/libc.so.6
#0  0x411722d6 in free () from /lib/libc.so.6
#1  0x415244e1 in operator delete () from /usr/lib/libstdc++.so.6
#2  0x415001ed in std::string::_Rep::_M_destroy () from /usr/lib/libstdc++.so.6
#3  0x41502d58 in std::basic_string<char, std::char_traits<char>, std::allocator<char> >::~basic_string () from /usr/lib/libstdc++.so.6
#4  0x414326e2 in Foctplot_command (args=@0x8bd1188, nargout=1)
    at octplot_command.cc:170
#5  0x403bd7bc in octave_builtin::do_multi_index_op (this=0x8968590,
    nargout=0, args=@0x8bd1188) at oct-obj.h:79
#6  0x403bcd15 in octave_builtin::subsref (this=0x8968590, type=@0x8c57d80,
    idx=@0xbfff80b0, nargout=1) at /usr/include/c++/3.3/bits/stl_list.h:678
#7  0x4039c137 in octave_value::subsref (this=0xbfff80d0, type=@0x8c57d80,
    idx=@0xbfff80b0, nargout=0) at ov.cc:837
#8  0x404a7f94 in tree_index_expression::rvalue (this=0x8c57d60, nargout=1)
    at oct-obj.h:79
#9  0x404a902d in tree_index_expression::rvalue (this=0x8c57d60)
    at pt-idx.cc:363
#10 0x4048a40e in tree_simple_assignment::rvalue (this=0x8d203f8)
    at oct-obj.h:51
#11 0x40489941 in tree_simple_assignment::rvalue (this=0x8d203f8, nargout=0)
    at oct-obj.h:51
#12 0x404d4d62 in tree_statement::eval (this=0x8cbd1a0, silent=false,
    nargout=0, in_function_body=false) at oct-obj.h:79

I played with the code a little, and it seems that just creating a temporary string_value to get the size of the input argument results in the segfault when the temporary gets deleted as it goes out of scope.

I tried setting GLIBCXX_FORCE_NEW=1 and rebuilding, but no change. Any other ideas?
Thanks in advance,

Paul Probert

[Shai Ayal]

Paul,

Can you explain your conclusion about the temporary string_value?

I see the error is in octplot_command.cc:170, and I don't see any string there.

Shai

p.s. GLIBCXX_FORCE_NEW=1 is a runtime parameter -- you don't have to recompile fot it to have effect.

[Nobody/Anonymous]

Thanks for your reply.
Line 170 in my editor is:
      char* str=new char[args(cur_arg+1).string_value().size()+1];

If I put a few lines ahead of it like this:

      std::cout << "testing..."<<std::endl;
      std::cout << "size="<<args(cur_arg+1).string_value().size()<<std::endl;
      std::cout << "end testing"<<std::endl;
      char* str=new char[args(cur_arg+1).string_value().size()+1];

Then I run it in octave:

octave:1> toggle_octplot
octave:2> line_demo
testing...
size=13
panic: Segmentation fault -- stopping myself...

So it faults before the next cout. Since the fault is in delete, I leapt to the conclusion that it is faulting when trying to deleted the temporary string_value that got created to call its size method.

Paul Probert

[Shai Ayal]

Thanks,

does changing it into a non temporary help? something along the lines of:

std::string tmp = args(cur_arg+1).string_value();
char *str=new char[tmp.size()+1];
strcpy(str,tmp.c_str());

Shai

[Nobody/Anonymous]

Thanks, Shai. I tried what you suggested. Even creating the std::string tmp, if I take the .size() of it I get the segfault.
  Something is very wrong with my setup here. I must have somehow gotten bad C or C++ libraries installed, probably sometime since I built octave. I have to go learn how to diagnose and fix this. I'm sure its nothing wrong with octplot.

Paul Probert