Hi Pierre,

probably my perception of what I saw led me to say "inconsistent". I will
explain:

My SystemC applications crash (depending on what kind of SystemC functions I
use) on CYGWIN and MINGW using either "quick threads" or fibers. They will run
on LINUX (quick threads) and Borland BCC free compiler (fibers). A minimum
SystemC example has been defined which causes a crash.

Within gdb the examples crash at certain points within the application. The
places where segmentation errors occur seem to be the same as if I ran outside
gdb. However SystemC is a relatively complicated assembly C++ classes.
Therefore I found it difficult to analyse its behaviour.

I ran a simple fiber example to learn about fibers and registers based on:
http://msdn.microsoft.com/library/devprods/vs6/visualc/vcsample/vcsmpfiberssamplefiberbasedfilecopyoperation.htm

This runs o.k. in CYGWIN or MINGW32 switching back and force between a read
and a write loop. Within gdb however after only two context switches it will
stop (not crash) with a return message greater than 0.

So I guess what I saw on SystemC was only half the message, and I am concerned
that without a proper debugger I will not be able to find the cause of the
SystemC crashes in CYGWIN and MINW32.

Holger Vogt


Pierre Muller schrieb:

> At 10:04 14/06/01 , Holger Vogt a écrit:
> >Hi,
> >
> >in CYGWIN or MINGW32 you may use fibers as "lightweight" threads
> >(defined in w32api). I tried to port SystemC (using fibers) to CYGWIN
> >but ran into difficulties.
> >
> >Using gdb to debug yields inconsistent results. A simple testcase exits
> >in gdb with error code returned but runs o.k. outside gdb (gdb 4.18 and
> >5.0).
> >
> >Are fibers supported in gdb? Is there a simple way to allow debugging
> >applications with fibers?
> This will probably be difficult,
> because the kernel API doesn't generate
> any debugger info when a fiber is created.
>
>    To support this, a special drivers that intercepts
> all these Fiber functions and calls the Debugger would be needed ...
>
>    Without it, GDB will continue to consider that you run a simple thread.
> But I don't really understand what you mean by saying that it yields
> inconsistent results.
>
>    Could you try to be a little more explicit ?
>
> Pierre Muller
> Institut Charles Sadron
> 6,rue Boussingault
> F 67083 STRASBOURG CEDEX (France)
> mailto:muller@ics.u-strasbg.fr
> Phone : (33)-3-88-41-40-07  Fax : (33)-3-88-41-40-99