Comment author: @oandrieu

callback(*solve_equation_closure, copy_string(eq))

Don't do this: IIRC the order of evaluation of the two operands is unspecified. solve_equation_closure need to be dereferenced after the call to copy_string().

value v=copy_string(eq);
return strdup(String_val(callback(*solve_equation_closure, v)));

Comment author: cniehaus

I tested it, and it crashes again :(

(I even made 'make clean && make' to be sure)

Perhaps it's necessary/better to apply this patch but it doesn't resolve the
crash

Comment author: cniehaus

This C-code creates the same crash. Much easier to track.

#include <stdio.h>
#include <stdlib.h>

int main(int argc, char ** argv)
{
· caml_startup(argv);

· char * eq = " a CH3(CH2)3COOH + b O2 -> c H2O + d CO2";
· char * eq2= " a CH3(CH2)3COOH + b O2 -> c H2O + d CO";
· char * result = solve_equation(eq);
· char * result2 = solve_equation(eq2);

· printf("solution : %s\n", result);

· printf("solution : %s\n", result2);
· free(result2);

· result2 = solve_equation(eq);
· printf("solution : %s\n", result2);
·
· free(result);
· free(result2);
·
· return 0;
}

Comment author: cniehaus

There is an easy way to reproduce this:

svn co svn://anonsvn.kde.org:/home/kde/trunk/KDE/kdeedu/kalzium/src/solver
make test
./testcprog

This crashes on at least three x64 machines I know, works on three of my x86 machines and on mellums alpha (as reported on IRC).

Comment author: @xavierleroy

Olivier Andrieu's comment about evaluation order is correct. You must fix your code as he suggests.

Moreover, you forgot to #include <caml/alloc.h>, so the compiler believes that copy_string returns an int instead of a long. This isn't the same thing on a 64-bit machine. Actually, gcc tells you just that:

ocamlopt -I +facile -c modwrap.c
modwrap.c: In function 'solve_equation':
modwrap.c:35: warning: implicit declaration of function 'caml_copy_string'