Re: [PATCH,v2] Make language setting tests more robust

[Luis Machado <lgustavo@codesourcery.com>]

On 02/06/2017 10:50 AM, Pedro Alves wrote:
> On 02/06/2017 02:54 PM, Luis Machado wrote:
>>
>> The story here is that i noticed 3 tests with such a problem:
>>
>> FAIL: gdb.compile/compile-ifunc.exp: nodebug: set language c
>> FAIL: gdb.dwarf2/data-loc.exp: set language ada
>> FAIL: gdb.dwarf2/dynarr-ptr.exp: set language ada
>>
>> And my build had a glibc with debugging symbols and sources, therefore
>> GDB had set the initial language to asm before main.
>
> There's a comment about that in gdb.compile/compile-ifunc.exp:
>
>     # This binary has no debug information, so force the language to C in case
>     # GDB happened to see real asm code when it stopped at the entry point
>     # when talking to a remote target, like gdbserver.  This guarantees the
>     # feature check below will work.
>     gdb_test_no_output "set language c" ""
>
> So the question is, why did the gdb_test_no_output work back then,
> and doesn't now?
>

I can't confirm it worked before since i haven't looked at this 
particular setup then.

> In top.c:check_frame_language_change()
> [...]
>   if (has_stack_frames ())
>     {
>       enum language flang;
>
>       flang = get_frame_language (frame);
>       if (!warned
> 	  && flang != language_unknown
> 	  && flang != current_language->la_language)
> 	{
> 	  printf_filtered ("%s\n", lang_frame_mismatch_warn);
> 	  warned = 1;
> 	}
>
> what does get_frame_language return?  Isn't it returning
> language_unknown for you?  Why not?
>

That returns language_asm for me. Supposedly because gdb has seen asm 
from glibc and then sticked with it when it reached the test program's 
main. And the program has no debug info.

>> FAIL: gdb.dwarf2/data-loc.exp: set language ada
>> FAIL: gdb.dwarf2/dynarr-ptr.exp: set language ada

Same as above, language_asm.

Maybe gdb should've switched to language_unknown when it reached main 
for this/these particular cases?

The logic to complain about a mismatch of language between the frame and 
the user selection seems sane.