Re: [Tinycc-devel] [PATCH v3] forbid invalid comparison of struct

[Hitoshi Mitake]

On Mon, Nov 5, 2012 at 11:04 PM, Thomas Preud'homme <address@hidden> wrote:
> Le lundi 5 novembre 2012 14:53:16, Hitoshi Mitake a écrit :
>> Thanks, I committed the change on the mob branch.
>
> Great.
>
>>
>> BTW, I've intended to implement comparison between structs
>> originally. And I found the bug accidentally.
>
> As I reviewed your patch and realized it's correct and I was wrong ;)
>
>>
>> Of course, comparison between different typed structs or structs and
>> scalar typed values should cause compile error. But I feel that if we
>> can write comparison between same struct typed values with ==
>> operator, a compiler will be very useful. Because we can avoid the
>> possibility of passing wrong value as 3rd parameter of memcmp().
>
> memcmp is not what you want because if a structure contains padding you are
> not sure of its content. Therefore two structures could be identical with
> different content inside the padding. You must compare field by field. That's 
> why
> assignement is possible but not equality: assignment is ok because you can
> copy the padding so memcpy will give a valid copy of the structure. Comparing
> with memcpy on the other hand will return false for some equal structures.

Ah, I missed about the padding... As you say, memcmp() is not a stuff for the
above situation.

>
>>
>> The standard of C doesn't allow this behaviour. But I think it may be
>> worth implementing it on TCC. Can TCC accept this behaviour as
>> implementation specific dialect?
>
> From http://bellard.org/tcc/ :
>
> UNLIMITED! Any C dynamic library can be used directly. TCC is heading torward
> full ISOC99 compliance. TCC can of course compile itself.
>
> I don't think we should not follow C99 on this.
>

OK, I understand the direction of TCC.
I'll implement == for field by field comparison only for my toy if I'll do :)

Thanks,

--
Hitoshi Mitake
address@hidden