Re: [patch] gdb/testsuite: fixes for gdb.xml/tdesc-regs.exp

[Sandra Loosemore <sandra@codesourcery.com>]

On 6/18/20 3:43 AM, Andrew Burgess wrote:
> * Sandra Loosemore <sandra@codesourcery.com> [2020-06-16 15:08:59 -0600]:
> 
>> On 6/16/20 2:47 PM, Andrew Burgess wrote:
>>
>>> I'd be interested to know more about which targets don't place any
>>> registers in the 'general' group.  This group is used in
>>> default_print_registers_info to implement 'info registers', so I'd
>>> like to see what this particular target has done instead.
>>
>> nios2, for one.  From the original test log for nios2-linux-gnu:
> 
> OK, but...
> 
> Please bear with me, I don't have a nios2 tool chain to hand, but...
> 
> In nios2-tdep.c, there is no call to set_gdbarch_print_registers_info,
> which means (from gdbarch.c) that nios2 will use
> default_print_registers_info.
> 
> Now if we look in infcmd.c at both default_print_registers_info and
> registers_info, then we see that if a user says:
> 
>    (gdb) info registers
> 
> then they will be asking which registers are in the general_reggroup.
> 
> Now, nios can optionally use a remote target-description (from
> nios2_gdbarch_init), so, lets for now assume no target description.  I
> see no call to set_gdbarch_register_reggroup_p for nios2, which means
> we are going to be using default_register_reggroup_p, which if we
> inspect (in reggroups.c) we'll see does place some registers into the
> general group.
> 
> Now, does this matter? The reggroup name is never printed anywhere, we
> just see a set of random registers?
> 
> But it is annoying that the user is not easily able to say:
> 
>    (gdb) info registers general
> 
> and get the same output.
> 
> If we look at how other targets deal with this most of them manually
> add at least some sub-set of the default register groups to their
> architectures set of register groups as part of their _gdbarch_init
> routine.
> 
> My gut instinct here is that this is what nios2 should be doing (and
> maybe arm too).
> 
> It does seem odd that there's no central "add-the-default-reggroups"
> type routine that targets can or should call.

OK.  How about I commit the other two pieces of the patch (which seem 
like genuine bugs in the testcase), and treat this one as uncovering a 
bug in the implementation instead of just an inappropriate x86-specific 
assumption wired into the testcase?

-Sandra