Among functions where we have almost no support for provider only keys is crypto/evp/p_lib.c, most of it relies solely on EVP_PKEY_METHOD and EVP_PKEY_ASN1_METHOD structures.

I just added EVP_PKEY_deassign() in the most evil of places: in the load_key() and load_pubkey() functions that all apps use. Tests are expected to go KABOOM.

Incidently, I looked into some of them that took me by surprise. test_pem is failing, and that caused me to look more closely at outputs... and some of them come out differently when printed via the provider serializers. This happens to be for invalid keys, but that still indicates that something's off, either with assumptions done when exporting from legacy to provider, or when serializing to text.

Fixups in (will be updated)

#10557, #10778, #10803, #10805, #10806, #10808, #10824, #10851, #10920, #10947, #11006, #11025, #11037, #11038, #11056, #11074, #11078

#11126 was a bit of a surprise... so it turns out that there's code in a few subsystems that are bypassing other central subsystems like EVP and ASN1 and call their backends directly, so weren't at all caught by this experiment (yet). CRMF seems to be one of them, I suspect CMS and a couple of others are similar. That means that we can expect to find a few "surprises" like that one...

All the PRs that were identified here are now merged, so a rebase was in order.

I still expects to see failures, but not quite as many. I also have another experiment in the making that specifically targets the sub-systems PKCS7 and CM*

Closing this PR, as it seems to have served its purpose. Some of the failures are still potentially there, but we do not need to do explicit deassignment of the legacy part of an EVP_PKEY to trigger them any more, as the key generation code is now serving us with such EVP_PKEYs and triggering code that hasn't been adapted for such keys yet.

Resurrecting this branch, to test #11422. Throw everything I have at it, basically...

Closing, it doesn't serve its purpose any more (or if it should, it needs to be rewritten entirely)