Re: [PATCH 2/2] gdb/testsuite: fix running gdb.linespec/cp-completion-aliases.exp with clang

[Andrew Burgess <aburgess@redhat.com>]

Andrew Burgess <aburgess@redhat.com> writes:

> Bruno Larsen via Gdb-patches <gdb-patches@sourceware.org> writes:
>
>> Currently, all important tests in gdb.linespec/cp-completion-aliases.exp
>> are failing when using clang because the wrong type is being suggested
>> for the completion.  For example, running with gcc and clang we get the
>> following output respectively:
>>
>> (gdb) break get_value(object_p) <- gcc version
>> (gdb) break get_value(object*)  <- clang version
>
> You are correct that what we print for Clang is not wrong.  But I don't
> think it's as correct as what we print for GCC.  The user wrote
> object_p, and the DWARF does encode the object_p for both versions.
> It's just in the Clang case there's an extra DW_AT_linkage_name which
> seems to send GDB off doing the "wrong" thing.
>
> I wonder how hard it would actually be to "fix" this so that we print
> object_p for Clang?  I think it would be helpful to really understand at
> which point we diverge when comparing Clang and GCC binaries.

The diff below fixes this issue for the specific test that you were
editing.  I've not tested this at all beyond that one test.  But I
wonder if something like this makes sense?

What's happening is that when we read the DWARF we end up calling
dwarf2_physname to figure out the name of the symbol.  If the DIE has a
linkage name then we just demangle that, and use that as the symbol
name.

But, at least for C++ I don't think that makes sense.  We know that the
demangled name will have stripped any typedefs.  While the computed name
will better reflect what's actually written in the code.

We already have something similar in place for Rust (commit
906bb4c58faa).

Anyway, just and idea...

Thanks,
Andrew


---

diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index ee4e7c1530a..00c46197e3c 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -9133,7 +9133,7 @@ dwarf2_physname (const char *name, struct die_info *die, struct dwarf2_cu *cu)
   if (!die_needs_namespace (die, cu))
     return dwarf2_compute_name (name, die, cu, 1);
 
-  if (cu->lang () != language_rust)
+  if (cu->lang () != language_rust && cu->lang () != language_cplus)
     mangled = dw2_linkage_name (die, cu);
 
   /* DW_AT_linkage_name is missing in some cases - depend on what GDB