I'm trying to locate FPEs in my own g77 lib, which works fine when I'm compiling statically. But when I compile my code into shared libs, I have problems finding the offending line.

Here's what I do for the static case:

- compile my lib code with -g

- link my main program against my lib and the "trapfpe" code mentioned in the g77 manual under "Trouble", "Missing Features", "Floating-point Exception Handling"

- load my main program into gdb, run it, and I get the offending line on a silver platter.


When I compile my lib into a shared object file, and link my main program against it, using trapfpe.so and LD_PRELOAD, it still aborts with "floating exception" alright, but running it in gdb simply produces output like this:

Program received signal SIGFPE, Arithmetic exception
0x41bbfc4d in ?? ()


What can I do to get the source line info in the case of shared objects too?

Thanks in advance for any pointers, I'm really stumped!

Stephan

The reason for the ()?? thing is that gdb has no symbol information.

In order to debug meaningfully you have to have linked -g.

You could simply keep a static library in your app until you've worked out all the issues.


It also sounds like you need to do some parameter checking before you execute the core math operations in your library. If something is out of range, find a way to report the problem.

Hi Jim,

thanks a lot for your reply, it's much appreciated!

I actually did this, my lib code is compiled with -g.

As long as I'm doing my own tests, yes. But the real problem is that this FPE occurs when my lib is linked to a commercial program, and that only takes shared libs (and is itself of course not compiled with -g, btw.).

True. For now I'm just looking for a way to locate the offending line.

But from your response I take it that what I'm trying do to at least should be possible, so I'm currently looking into other potential reasons. For instance, I might have picked up a non-debug version of my lib because I didn't update LD_LIBRARY_PATH properly...

All the best,
Stephan

If you variable really holds a float point value and you do something like fp := (fp*100.00)/100.00;
than the range of the float point value is changed and you get that signal. So trunc or round that floatpoint before the range will be changed. It's a CPU-fault like division by 0.