Re: [RFA] 12266 (typedef'd parameters) revisited again - what should go for gdb-7.3?

[Jan Kratochvil <jan.kratochvil@redhat.com>]

On Sun, 05 Jun 2011 15:28:15 +0200, Jan Kratochvil wrote:
> On Sat, 04 Jun 2011 01:33:26 +0200, Keith Seitz wrote:
> > Of course, it is still possible that I've neglected something along
> > the line.
> 
> This patch crashes for me on:

You have said off-list you have a fix but you still run the real-world .debug
files regression tests on top of it.

I believe the replace_typedefs part (DEMANGLE_COMPONENT_* typedef
replacements) should be cross-checked against `set debug check-physname on'
which I guess hasn't been done yet.

This patch fixes on its own PR 12506 stated as a blocker for:
	http://sourceware.org/gdb/wiki/GDB_7.3_Release

The question (at least for me) is now which parts of this patch should go for
7.3 and which only for 7.4+:

It can be split into the parts before before:
1/4: changes cp-name-parser.y to not use shared memory.
2/4: differences between print names and "linkage names" (physnames).
3/4: cp_canonicalize_no_typedefs to decode_variable and decode_compound
4/4: tests for this new feature.

PR 12506 is I believe fixed by 2/4.  But for compilers featuring
DW_AT_MIPS_linkage_name (=G++) it is fixed also by the more general:
	Re: [patch] Follow DW_AT_linkage_name for methods #2
	http://sourceware.org/ml/gdb-patches/2011-06/msg00040.html
And I believe I could still find a countercase where DW_AT_linkage_name is
more correct than the current physname code.

ctors/dtors do not have DW_AT_MIPS_linkage_name but for class based naming
(C::C or C::~C) one can drop to the minimal symbols (the fallback needs to
work even for other cases):
	[patch 1/2] physname reg.: linespec minsym fallback
	http://sourceware.org/ml/gdb-patches/2011-06/msg00079.html
and one IMO cannot use linespec to refer to ctors/dtors (only to regular
methods) during variable base naming (var.C or var.~C) - where the minsym
fallback is not applicable and where decode_compound could be useful.

The 3/4 is a new feature for PR 12266.  GDB could never
do `break f(typedefparam)', I do not think it needs to go for 7.3 so late.


Thanks,
Jan