Re: [gdb-7.1] 10 days to branching...

[Pedro Alves <pedro@codesourcery.com>]

On Friday 05 February 2010 22:02:22, Tom Tromey wrote:
> >>>>> "cgf" == Christopher Faylor <cgf-use-the-mailinglist-please@sourceware.org> writes:
> 
> cgf> Any possibility that you can confirm whether this is a windows-only
> cgf> problem?
> 
> cgf> I think the only way to do that is to provide a test case.
> 
> It is not Windows-specific.  I reproduced it with a c++ program on my
> x86 F11 box.
> 
> The program comes from PR 9032:
> 
>     #include <stdio.h>
>     class InstSelection {
>     public:
>       InstSelection(double d) {
>       }
>     public:
>       virtual ~InstSelection(void);
>       void dump(void);
>     };
> 
>     InstSelection::~InstSelection(void)
>     {
>     }
> 
>     void InstSelection::dump(void)
>     {
>       printf("InstSelection::dump here\n");
>     }
> 
>     int main(void)
>     {
>       InstSelection *is = new InstSelection(17.0);
>       is->dump();
>       delete is; is = 0;
>       return 0;
>     }
> 
> I compiled it with the system g++, then ran CVS head gdb on the
> executable.  Finally:
> 
> (gdb) b InstSelection::InstSelection
> ../../src/gdb/breakpoint.c:4962: internal-error: set_raw_breakpoint: Assertion `sal.pspace != NULL' failed.
> 
> 
> The sal causing the problem in set_raw_breakpoint:
> 
> (top-gdb) p sal
> $10 = {
>   pspace = 0x0, 
>   symtab = 0x0, 
>   section = 0x0, 
>   line = 0, 
>   pc = 0, 
>   end = 0, 
>   explicit_pc = 0, 
>   explicit_line = 0
> }
> 
> That seems odd.

This one's been described and diagnosed in
PR10966 <http://sourceware.org/bugzilla/show_bug.cgi?id=10966>

The expr-cumulative work fixes this, but I listed in the PR
two possible quick workarounds.

Has a small objc testcase been found yet?  It could
simply be something forgetting to set the sal's pspace
around decode_objc, but then I'm confused at how it only
triggers in some cases, and the objc tests in our
testsuite pass.  It could yet be another bug that
was masked out before the pspace assertion was added,
like the PR10966 bug.

-- 
Pedro Alves