Re: [patch] testsuite/gdb.c++/local.exp: accept more nested types

[Daniel Jacobowitz <drow@mvista.com>]

On Mon, May 27, 2002 at 01:27:53PM -0500, Michael Elizabeth Chastain wrote:
> Daniel Jacobowitz writes:
> 
> mec> This patch updates local.exp to track an improvement in gdb.
> mec> In a configuration with gcc HEAD -gstabs+, gdb prints nested types
> mec> as "InnerLocal::NestedInnerLocal" instead of just "NestedInnerLocal".
> 
> dj> It's an improvement in GCC's debug output, I believe; I seem to recall
> dj> Kevin posting a patch for this to gcc-patches recently.
> 
> I don't think it's a change in gcc because gcc does not show this change
> with gdb 5.2 or gdb gdb_5_2-branch.  It happens only with gdb HEAD.
> 
> Although there have been changes in gcc this week in this area.  Perhaps
> it really is a gcc change, and gdb HEAD is the only gdb looking at the
> area of the change.

Let's go with a working philosophy of "both" then :)

> dj> Is it really?  It seems to me that we should strip off the name of the
> dj> enclosing class from nested types.  This becomes more of an issue with
> dj> the DWARF-2 type printing I'm experimenting with.
> 
> This is a philosophical issue.  Let's explore it.

Indeed it is.

> gdb's output is:
> 
>   # target=native, host=i686-pc-linux-gnu%rh-7.2, gdb=HEAD%20020526
>   # gcc=HEAD%20020526, binutils=HEAD%20020526, glibc=VENDOR,
>   # goption=-gstabs+
> 
>   ptype InnerLocal
>   type = class InnerLocal {
>     public:
>       char ilc;
>       int *ip;
>       InnerLocal::NestedInnerLocal nest1;
> 
>       InnerLocal & operator=(InnerLocal const&);
>       InnerLocal(InnerLocal const&);
>       InnerLocal();
>       int il_foo(unsigned char const&);
>   }
> 
> gdb presents me with this output.  What result should I assign to it?
> 
> I think this is a PASS.  I think that the output is clear to humans.
> The construct "InnerLocal::NestedInnerLocal nest1" is also valid C++;
> if you add back in the definition of NestedInnerLocal, then it will
> compile.
> 
> I also accept "NestedInnerLocal nest1;" as a PASS.
> 
> Philosophically, I'm not interested in the "one true output".  I look
> at what gdb actually prints in each configuration, and then classify
> these into PASS and FAIL.  This is especially true for "ptype" output,
> where the output depends on the compiler and debug format.
> 
> What do you think?

I think that I (as C++ maintainer rather than testsuite maintainer)
need to figure out what the correct output is. 
"InnerLocal::NestedInnerLocal nest1" is not actually valid C++ in
general, I don't believe; what if there is a class
InnerLocal::InnerLocal::NestedInnerLocal?  I believe it will resolve to
that one instead of the correct one.  This is also uglier; for std::
types there's going to be a lot of useless noise.

The ideal output, it seems to me, would be something like:

>   ptype InnerLocal
>   type = class InnerLocal {
>     public:
>       char ilc;
>       int *ip;
>	class NestedInnerLocal;
>       NestedInnerLocal nest1;
> 
>       InnerLocal & operator=(InnerLocal const&);
>       InnerLocal(InnerLocal const&);
>       InnerLocal();
>       int il_foo(unsigned char const&);
>   }

Unambiguous, quite obviously correct.  By this logic,
InnerLocal::NestedInnerLocal deserves a KFAIL comment for the moment,
since the code to do this is not yet present.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer