Re: [RFA/DWARF] Add DW_AT_GNAT_descriptive_type support

[Joel Brobecker <brobecker@adacore.com>]

> People developing non-Ada GDB code sometimes accidentally cause Ada
> regressions but the current Ada debug info (at least as encoded in DWARF) is
> unreadable for me, despite I tried.
> 
> Is there some Ada-specific `readelf -wi'-like tool to be able to read it?

Unfortunately not. In order to understand the debugging info, you need
to read the GNAT encoding spec in exp_dbug.ads (in the compiler
sources).

> And why not to implement whole Ada properly in DWARF, with few extensions
> proposed at <dwarf-discuss@lists.dwarfstd.org> when/if needed?  As AdaCore
> supplies both the compiler and the debugger I see no need to be bound by STABS
> you noted before.

We simply do not have the resources, and we don't project to have these
resources any time soon. Also, we are still stuck with stabs for a while,
as we haven't been able to transition to DWARF on some of the platforms
that AdaCore supports (Eg: AIX, Tru64, some versions of LynxOS, HP/UX). 
We've identified some encodings that should be easy to express in pure
DWARF, and these changes are in the queue, but the rest won't be done
any time soon, I'm afraid.

> As Ada regressions are AFAIK not acceptable for FSF GDB check-ins
> there should be a way to make them troubleshootable by non-Ada
> developers.

I only expect the contributors to do their best: If they notice
some regressions in gdb.ada with their patch, then try seeing if
it's an obvious problem, but otherwise just ask either Paul or myself.
I'm completely overworked these days, but I will always make room
to help out. I understand that the Ada part can be obscure, so
I don't expect more than that. And if someone introduces a regression,
then oh well, I'll notice it eventually, and then go on to fix it.

-- 
Joel