Temporarily add pragmas

[bjwtaylor]

Description

Temporarily add pragmas to allow switching of function prototypes contributes to Mbed-TLS/TF-PSA-Crypto#346.

This PR is part of a chain which needs to be merged in the following order:

1. Temporarily add pragmas #10271
2. Create new enum mbedtls_pk_sigalg_t TF-PSA-Crypto#356
3. Update functions for new mbedtls_pk_sigalg_t #10290
4. Revert Pragmas TF-PSA-Crypto#377

PR checklist

• changelog provided | not required because: TBC
• development PR provided #HERE
• TF-PSA-Crypto PR provided #Create new enum mbedtls_pk_sigalg_t TF-PSA-Crypto#356
• framework PR not required
• 3.6 PR not required because: No backports
• tests provided

[gilles-peskine-arm]

As discussed in Slack, this is not acceptable. It'll hide errors for everyone until we undo it.

I don't know what the problem is precisely. It would be ok to disable a specific error locally with a #pragma GCC diagnostic (grep for examples where we know that some warning is spurious). Or maybe we should do the transition differently.

[gilles-peskine-arm]

How about the following transition, would this work?

1. In crypto, make mbedtls_pk_sigalg_t an alias of mbedtls_pk_type_t.
2. In mbedtls, replace mbedtls_pk_type_t by mbedtls_pk_sigalg_t (so we'd be doing most of Switch to mbedtls_pk_sigalg_t #10264 before Create new enum mbedtls_pk_sigalg_t TF-PSA-Crypto#356).
3. In crypto, change mbedtls_pk_sigalg_t to be the desired type.
4. In mbedtls, finish Switch to mbedtls_pk_sigalg_t #10264.

[bjwtaylor]

I've updated the chain to use pragma's, which I think is the easiest way to get the changes in. Let me know what you think though?

[valeriosetti]

The #pragma change is OK to me, but putting it on top of a C file then it applies to it all. I know that this change is temporary and since there's no other change in this PR we should be fine, but I would prefer to limit this "disabling" only on where it's strictly needed inside the C file. I think you can use:

#pragma GCC diagnostic push
#pragma GCC diagnostic warning "-Wenum-conversion"
....
#pragma GCC diagnostic pop

to store/resume checks around the problematic lines. Do you think this is feasible?

[gilles-peskine-arm]

The pragmas need to be protected from compilers that don't recognize them.

I think it would be easier to write a single conversion function and use that, rather than put pragmas everywhere.