Re: [PATCH v2 4/8] DWARF-5 basic functionality

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

On Tue, 21 Feb 2017 20:18:36 +0100, Andreas Arnez wrote:
> When I compile with "-O3", GCC now yields warnings for the two changed
> lines above:
> 
>   [...] warning: ‘header.comp_unit_head::type_offset_in_tu.cu_offset::cu_off’
>   may be used uninitialized in this function [-Wmaybe-uninitialized]

What compiler/version/options do you use?  I have tried now
	gcc-4.8.5-11.el7.x86_64
	CFLAGS=-O3 CXXFLAGS=-O3 ./configure;make

While I get some warnings/errors none of them is in dwarf2read.c.

I do not want to patch it without reproducibility of the warnings.


> I *think* GCC is wrong.  From what I understand, we can only get here if
> header.unit_type == DW_UT_type, and then these fields should have been
> initialized in read_comp_unit_head before (right?).

Yes, I hope so.


> But then I wonder
> about the effect of the call to create_debug_type_hash_table in
> create_all_type_units:
> 
>   create_debug_type_hash_table (NULL, &dwarf2_per_objfile->info, types_htab,
> 				rcuh_kind::COMPILE);
> 
> Is that needed?

Yes.  Otherwise GDB would not read definitions of classes during
	-gdwarf-5 -fdebug-types-section
as then DW_UT_type are located inside .debug_info.


> When doing so (see untested patch below),

No, that breaks: -gdwarf-5 -fdebug-types-section


Jan