Remove double underscores from macro and add a check for it

[AndrzejKurek]

This PR addresses issue #171.
It removes the double underscore from a macro, adds an exception for it in "test_psa_constant_names", and introduces a check for double underscores in check-names.sh.

[Patater]

LGTM

Check names commit should also go to Mbed TLS development branch (no other ports needed)

[dgreen-arm]

LGTM

[athoelke]

This will now result in PSA_ALG_AEAD_WITH_DEFAULT_TAG_LENGTH_CASE appearing in the PSA Crypto specification as all symbols that begin with PSA_ALG_ are considered part of the architectural specification rather than part of the mbed implementation that should not be part of the interface spec.

Although we could fix this with a matching change to the tools that extracts the spec documentation from the mbed implementation (which we still do for the moment), it is not an ideal situation.

Ideally, there needs to be some reserved name spaces defined for the PSA specification, such that implementations would be able to allocate symbols for their own internal and extension APIs without risk of collision with the specification. It may be late to request that mbed not use a psa_ or PSA_ prefix for its own APIs - although that is likely to be what that specs will recommend to avoid clashes with future version of the specification.

[gilles-peskine-arm]

Ideally, there needs to be some reserved name spaces defined for the PSA specification, such that implementations would be able to allocate symbols for their own internal and extension APIs without risk of collision with the specification.

I agree. Does the PSA framework define this?

It may be late to request that mbed not use a psa_ or PSA_ prefix for its own APIs - although that is likely to be what that specs will recommend to avoid clashes with future version of the specification.

We've been inconsistent about this. The first time I added an implementation-specific function to Mbed Crypto, I called it mbedtls_psa_xxx. But then I didn't follow this convention for other implementation-specific functions which I considered to be experimental, possibly to be made official at a later date. For macros, I never used MBEDTLS_PSA_xxx because Mbed Crypto has a very fragile script that checks for possible typos in preprocessor symbols that begin with MBEDTLS_ and I wanted to minimize PSA-related modifications, but I think now we can move past that (#241).