[cfe-dev] debugging clang driver backtrace

[Rafael Espíndola]

Running gdb on the "clang -cc1" command is probably the easiest. You
can use gdb's "bt" to print a stack trace once you get to the assert.

On 26 October 2012 14:47, Brian Brooks <brooks.brian at gmail.com> wrote:
> Hello cfe-dev!
>
> I am taking a stab at clang bug id 14085.
>
> I put the offending source in tmp.cc and started off with "gdb clang++" and
> then "(gdb) run tmp.cc".  This produces an "UNREACHABLE executed at ...."
> with a really helpful backtrace.
>
> When I break at main() and step through the driver code, a posix_spawn() is
> eventually executed, and I am not sure how to step through that.  So, I
> figured it may be better to debug clang with the extensive command line
> arguments that are generated (this is dumped into a .sh file and recommended
> you attach it when submitting bug report).
>
> So I "gdb clang" and "(gdb) run -cc1 -triple x86_64-unknown-linux-gnu
> ......... tons of options ....... tmp.cc" and break at cc1_main().  When I
> continue, it hits the same "UNREACHABLE executed at ....", but the backtrace
> is not printed.
>
> Can someone explain why this is?  Is the backtrace dumped somewhere that the
> driver pid can read and that is why I see it in the first debugging
> scenario?
>
> Any other recommendations for someone getting started with debugging clang?
> Or, how to debug through the posix_spawn() in the driver?
>
> Thanks!
> Brian
>
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>