Re: [PATCH] gdb/i387-tdep.c: Avoid warning for "-Werror=strict-overflow"

[Pedro Alves <palves@redhat.com>]

On 10/04/2014 06:18 AM, Chen Gang wrote:
> On 10/4/14 1:49, Pedro Alves wrote:
>> On 10/03/2014 05:44 PM, Joel Brobecker wrote:
>>>>> Sorry, but obfuscating code to make compilers happy is *not* the way to go.
>>>>>
>>>>
>>>> OK, I can understand, but for me, these is no other better ways for it,
>>>> except let gdb give up "-Werror" (if always need "--disable-werror"
>>>> during "configure").
>>>
>>> I have to agree with Mark on this one, the proposed solution looks
>>> awful. There has to be another way. Maybe declaring a local constant
>>> whose value is I387_XMM0_REGNUM (tdep)?
>>
>> Likely, after transformations and intra-procedural analyses, gcc would
>> end up with the same.
>>
>> This:
>>
>>  for (i = I387_ST0_REGNUM (tdep); i < I387_XMM0_REGNUM (tdep); i++)
>>
>> always iterates exactly 16 times, because I387_XMM0_REGNUM
>> is defined like:
>>
>>  #define I387_XMM0_REGNUM(tdep) (I387_ST0_REGNUM (tdep) + 16)
>>
>> An alternative I think might work would be to give that magic
>> 16 constant a name, say:
>>
>>  #define I387_NUM_ST_REGS 16
>>
>> and then do:
>>
>>  for (i = 0; i < i < I387_NUM_ST_REGS; i++)
>>    {
>>       int r = I387_ST0_REGNUM (tdep) + i;
>>
>>       ... use 'r' instead of 'i' ...
>>    }
>>
> 
> OK, thanks. It is really one way, it is a little better than my original
> way. But for me, it is still not a good idea: it introduces a new macro
> and a new variable for each area (originally, it is only one statement).

I see no problem with adding the new macro.  We already have a ton
of similar macros, see i386-tdep.h and i387-tdep.h.  Looks
like the existing I387_NUM_REGS is what we'd need here?

BTC, OOC, did you try Joel's idea with the local variable?
In case Mark prefers that, it'd be good to know whether it works.
I can't seem to get my gcc to emit that warning.

Combining both ideas, for clarity, we end up with something
like:

 int end;

 end = I387_ST0_REGNUM (tdep) + I387_NUM_REGS;
 for (i = I387_ST0_REGNUM (tdep); i < end; i++)

 ...

 end = I387_XMM0_REGNUM (tdep) + I387_NUM_XMM_REGS (tdep);
 for (i = I387_XMM0_REGNUM (tdep); i < end; i++)


That's way clearer to me than the existing:

 for (i = I387_ST0_REGNUM (tdep); i < I387_XMM0_REGNUM (tdep); i++)
...
 for (i = I387_XMM0_REGNUM (tdep); i < I387_MXCSR_REGNUM (tdep); i++)

anyway, which assumes the reader knows register numbers are
ordered like st -> xmm -> mxcrsr.

If this works, I think it's my preference.

> For me, "-Werror" need always be optional, but not mandatory.

It's mandatory only on development builds.  -Werror is not on by
default on released GDBs.

Thanks,
Pedro Alves