This is the mail archive of the
gdb@sources.redhat.com
mailing list for the GDB project.
Re: C++/Java regressions
On Wed, Nov 26, 2003 at 04:04:55PM -0500, Ian Lance Taylor wrote:
> Daniel Jacobowitz <drow@mvista.com> writes:
>
> > On Tue, Nov 25, 2003 at 11:04:41PM -0500, Ian Lance Taylor wrote:
> > > Then when gdb goes to look up T5<int>, it finds the DWARF psymbol for
> > > the class itself, but it also finds that the symbol might be a
> > > namespace. It then decides that it is a namespace.
> >
> > That sounds like the bit that's wrong. If it found a psymbol for the
> > class why does it decide to call it a namespace?
>
> Good question. My guess is that it's because lookup_symbol_aux()
> calls current_language->la_lookup_symbol_nonlocal() before it calls
> lookup_symbol_aux_psymtabs(). If I force la_lookup_symbol_nonlocal()
> to return NULL, then lookup_symbol_aux_psymtabs() finds the typedef,
> and `ptype T5<int>' works more or less correctly.
>
> One solution might be that when gdb finds that a symbol has a class
> definition, it makes sure that it does not have a pseudo-namespace
> definition.
The pseudo-namespaces are always supposed to be shadowed by symbols.
>From your explanation it looks like they're shadowing psymbols; but
since psymbols are supposed to behave identically to symbols (that's
the whole point), something is wrong.
I'll look at it if I get a chance before David Carlton does; this is
originally his code. Thanks for the investigation.
> > I'm considering a routine for canonicalization of C++ demangled names.
>
> If it helps, that's more or less what my libiberty C++ demangler does.
> It first translate the name into a simple tree structure, and then
> walks the tree translating it into a string. I could expose the tree,
> although it would have to be documented a bit better.
Oh, so it's already two-pass? If we could work out an API for the
tree, then GDB could build and supply trees to get a canonical form
back from the demangler. That has the added bonus of not needing to
post-process the demangler output (if we know that we've got a GNU v3
name and thus this demangler was used, of course - v2 would still need
to be parsed). That sounds like an ideal solution.
Do you have an opinion on the "A::bar const" DMGL_PARAMS issue that
Michael Chastain noticed, by the way?
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer