sha1 hash using libgcrypt different from what returns sha1sum

[Jussi Kivilinna]

On 14.12.2013 03:37, Rafaël Carré wrote:
> 
> 
> Hello,
> 
> Werner Koch <wk at gnupg.org> a écrit :
>> On Fri, 13 Dec 2013 21:51, jussi.kivilinna at iki.fi said:
>>
>>> That's right, size_t for lengths is not enough on 32-bit. Could we
>> use uint64_t
>>> instead of size_t?
>>
>> uint64_t is not really portable because C99 is not deployed everywhere.
> 
> Are you thinking about Visual Studio and its incomplete C99 support?
> 
> I hope that's the only compiler left that doesn't support C99. And I'm not sure you should keep gcrypt off of 15 years old features just because of it.
> 
> If there're others compilers I'd like to know because I've been using C99 extensively for a few years.
> 
> In any case we've had these types in mingw for a while and I think there are copies of a MSVC compatible stdint.h in the wild.
> 
>> However, this is a minor issue because we could enable CCM only if we
>> have that type.  We do the same for some algorithm (TIGER comes to
>> mind).
>>
>> We could document a certain limit on CCM or we look for a more general
>> solution.  In Windows this has traditionally be solved using unions and
>> structs.  It's not nice API, though.
>>
>> off_t would be a natural choice but it has problems as well.  For
>> example the two defined ABI variants and I have not checked whether it
>> is suitable for a size_t replacement (I guess not).
> 
> I think off_t depends on __USE_FILE_OFFSET64 at least on 32 bits so it doesn't sound like a good choice.
> 
>> What do we we need to change in the API to correctly support it.  What
>> side-effects will be have if we use uint64_t and provide the interfaces
>> only if uint64 is defined?
> 
> Applications failing to link correctly when built with the non C99 able compilers I guess, I can't think of something else.

With GCRYCTL_SET_CCM_LENGTHS, linking should not be problem since CCM data
lengths are passed in through:

gcry_error_t gcry_cipher_ctl (gcry_cipher_hd_t h, int cmd, void *buffer,
                             size_t buflen);

So uint64_t would not be used for any function prototype or any types. When
libgcrypt is build without uint64_t, GCRYCTL_SET_CCM_LENGTHS would return
error.

-Jussi

> 
>> Shalom-Salam,
>>
>>   Werner
> 
> Regards,
> 
> 
> _______________________________________________
> Gcrypt-devel mailing list
> Gcrypt-devel at gnupg.org
> http://lists.gnupg.org/mailman/listinfo/gcrypt-devel
>