PSA return status coverage script

[gilles-peskine-arm]

Add infrastructure to run unit tests and collect the return values for every PSA API function that returns psa_status_t.

./tests/scripts/psa_collect_statuses.py >statuses.txt

Status: runs ok but not very well documented and the build script is not very robust.

[Patater]

This may conflict with #213

I like the idea of checking that return codes are documented. I think the API is stable enough now that having that capability will be a help and not a hindrance.

Is it possible to continue work on this?

[gilles-peskine-arm]

I'll rebase this on top of #213 if we decide to adopt it.

Are you ok with the current design? It was good enough for my goal at the time, which was to compare reachable errors with documented errors and is still relevant. But it isn't very well integrated into Mbed Crypto. The script I have for the other side, to collect documented errors, works from the PSA spec (private link: https://github.com/ARMmbed/psa-crypto/pull/227).

[Patater]

Are you ok with the current design?

It's a bit funky in that its a mix of shell in a Makefile and some Python, but it's beneficial as is. We can refactor into something better as needed.

[gilles-peskine-arm]

I rebased the changes on top of current development. I added some documentation, but I did not change the design. I think this is good enough to include, and we'll see later whether to improve the design or to leave it as a minor hack depending on how useful we think this is.

[Patater]

LGTM

[Patater]

We may not need to add unset MBEDTLS_MEMORY_BUFFER_ALLOC_C if we sideport the removal of MBEDTLS_MEMORY_BUFFER_ALLOC_C from full config. Necessary for now.

[gilles-peskine-arm]

We definitely should sideport that change. I'd expect it to come around the next time we sync from mbedtls.

[k-stachowiak]

I agree it's a bit "funky", but all the tools make sense on their own, which makes them fine by me. I made a comment regarding the build logs. If the internal build fails on a remote machine, it may be difficult to learn the failure reason. I had such a failure locally, and had to manually call make test to see, what was wrong. Otherwise, LGTM.