Fix pointer to AEAD context

[silkeh]

The crypto_aead functions with precomputation take a pointer to a
crypto_aead_aes256gcm_state object as argument, not a char array.

The compiler is changed to clang because the code with the correct pointer does
not compile with GCC, see golang/go#7270.

[redragonx]

Does it compile on Windows too?

[silkeh]

I don't see why not, but I don't have any way to verify.

[redragonx]

Hmm, ok.

Me neither, the batch file needs updating before I can accept this PR. Maybe it's a small change.

[silkeh]

I have changed the batch file as well.

[theodelrieu]

Is there any news on this? It doesn't compile on macos with apple-clang 9.0 either, this patch fixes it

[theodelrieu]

It's weird that it breaks GCC, crypto_aead_aes256gcm_state is just a C typedef. This PR fixed clang but broke GCC.

Do you think there is a way to make both happy?

[jedisct1]

That structure must be stored at a 16 bytes aligned location.

The same thing applies to crypto_onetimeauth_poly1305_state.

[silkeh]

@theodelrieu GCC has an open issue for this: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81001

[jedisct1]

This seems to explain how to deal with alignment constraints: http://www.tapirgames.com/blog/golang-memory-alignment

[silkeh]

@jedisct1 I will look into alignment constraints for this (and for #31), thanks!

[redragonx]

Thank you to everyone that's helping with this issue.

[silkeh]

I have updated #31 and created #32 for the alignment issue.

[redragonx]

Does this work for CLANG and GCC now then?

[silkeh]

@redragonx no, that is not possible until this GCC bug is solved.