Re: [PATCH, updated] Add support for setting disassembler-options in GDB for POWER, ARM and S390

[Pedro Alves <palves@redhat.com>]

On 02/14/2017 05:21 PM, Yao Qi wrote:
> On Mon, Feb 13, 2017 at 4:58 PM, Pedro Alves <palves@redhat.com> wrote:
>> On 02/13/2017 03:52 PM, Yao Qi wrote:
>>> These options should be modeled as per-architecture data.  We need to
>>> define a key to access that data dynamically.  grep
>>> "static struct gdbarch_data *" in *.c.
>>
>> If I understand the suggestion correctly, that would make all the different
>> POWER (etc.) gdbarch instances have their own instance of the option string.
> 
> Hi Pedro,
> Peter's patch makes each gdbarch instance has its own instance of
> option string set by GDB.  Peter's patch adds two things in gdbarch,
> 
> +# Functions for allowing a target to modify its disassembler options.
> +v:char *:disassembler_options:::0:0::0:pstring (gdbarch->disassembler_options)
> +v:const disasm_options_t
> *:disassembler_options_arch:::0:0::0:host_address_to_string
> (gdbarch->disassembler_options_arch->name
> 
> The former one is a string, and it is per-gdbarch-instance.  The latter is a
> function, which provides all valid options for an arch.  Different gdbarch
> instances for powerpc, for example, have the same valid options, but each
> gdbarch instance will have its own setting (set by GDB command).

Then I'm a bit surprised, since that was not what was discussed
earlier, here <https://sourceware.org/ml/gdb-patches/2016-11/msg00043.html>:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> I didn't realize we could have multiple gdbarchs for the same
> arch live at the same time.  Do you have a suggestion on how
> to fix this?  Maybe each arch will have its own *-tdep.c global
> var to hold it and maybe store a pointer to that in the gdbarch?

Yeah, that sounds like the simplest way to handle it.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

In order to implement the above, the gdbarch should be storing
a "char **", but it is storing a "char *" instead:

  +v:char *:disassembler_options:::0:0::0:pstring (gdbarch->disassembler_options)

So I don't know why the current patch is doing it differently,
and I was going by that earlier discussion.

> My suggestion about using "struct gdbarch_data *" doesn't change the
> the behaviour.

Thanks for checking.  So the question now is why does the
current patch want to behave like that.  :-)

Thanks,
Pedro Alves