Re: [PATCH] sparc: Remove optimize md5, sha256, and sha512

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 17/07/23 14:31, Sam James wrote:
> 
> Adhemerval Zanella Netto <adhemerval.zanella@linaro.org> writes:
> 
>> On 17/07/23 12:01, Sam James wrote:
>>>
>>> Adhemerval Zanella via Libc-alpha <libc-alpha@sourceware.org> writes:
>>>
>>>> The libcrypt was maked to phase out on 2.28, and better projects
>>>> already exist that provide both compatibility and better API
>>>> (libxcrypt).  The sparc optimizations add the burden to extra
>>>> build-many-glibcs.py configurations.
>>>>
>>>
>>> I guess. I'd personally prefer just removing it all when libcrypt
>>> is removed, not hacking away at it beforehand.
>>
>> I don't think we will ever be able to remove libgcrypt due compat reason.
>> Maybe we can make crypt return ENOSYS as default, but even though I think
>> if libcrypt is still enabled we will need to keep old code as compat
>> layer. 
>>
> 
> In the other thread, we're talking about when it's time to kill it
> entirely (I suppose just the option to build, not the real
> implementation).

I am not sure what kind of compatibility we will need to preserve for
libcrypto.  For sunrpc 5500cdba401), we have removed most of the symbols
but still kept some old ones for compatibility even though the
libtirpc does implement such symbols (like key_setsecret for instance).

It seems that libcrypto is way simpler, and libxcrypt does implement
all require symbol version which the expected. So it should be doable.

> 
> My assumption was we'd remove optimisations like this at that point.

The idea of the patch is to streamline the required build coverage to
avoid having a --enable-crypt just for sparc.  Sure we can pack if/when
we remove libcrypto, but I think this is orthogonal.