Re: Another update of GNU TLS bindings

[Simon Josefsson]

"Stefan Monnier" <monnier+gnu/address@hidden> writes:

>> * GC fixes as suggested.  However, Lisp_Process in process.h still
>>   contain #ifdef's (and thus cause a binary API incompatibility)
>>   because the type of the newly added variables is not known unless
>>   TLS is enabled.  Is this OK?  I'm not sure how this could be solved.
>
> You could give a default bogus definition for those types like
>
> #ifndef HAVE_GNUTLS
>   typedef void *GNUTLS_STATE;
>   typedef void *GNUTLS_X509PKI_CLIENT_CREDENTIALS;
>   typedef void *GNUTLS_ANON_CLIENT_CREDENTIALS;
>   typedef void *GNUTLS_SRP_CLIENT_CREDENTIALS;
> #endif

I considered this, but then recalled that maybe the C standard doesn't
guarantee binary compatibility between different pointer types so it
wouldn't be guaranteed to be better.  (Even if this isn't the case,
the GNUTLS types aren't guaranteed to be pointers at all, but we
already cheat during initialization, so...)

But I guess in practice it works, so I changed it, thanks.