make sure we keep testing OSSL_QUIC_client_thread_method clients

[Sashan]

This snippet comes from do_test() function found test/quic_tserver_test.c :

157     if (!BIO_dgram_set_peer(c_net_bio, s_addr_))
158         goto err;
159 
160     if (!TEST_ptr(c_ctx = SSL_CTX_new(use_thread_assist
161                           ? OSSL_QUIC_client_thread_method()
162                           : OSSL_QUIC_client_method())))
163         goto err;
164 
165     if (!TEST_ptr(c_ssl = SSL_new(c_ctx)))
166         goto err;

There we create a client context which is used to connect to t-server. This particular piece needs to be salvaged from the quic_tserver._test.c before it will get scrapped to prevent losing test coverage for thread assisted tests.

I think it's worth to investigate an option if it would be possible to add support for thread assisted clients to radix tests. The testing needs to cover those situations:

• SSL_read()/SSL_write() works
• the connection close is handled properly
• stream reset is handled properly

may be it will turn out this deserves stand alone test if we will find no easy way to plug it it in into radix..

also it looks like the current testing in quic_tserver does cover only default stream mode. But we should perhaps extend coverage for SSL_stream_new()/SSL_stream_accept() too.

[bukka]

Interestingly I was just looking into this as a follow up to the openssl/openssl#31746 (review appreciated :) ). I'm still getting my head around RADIX but it seems to me like it will have similar issue with this race. So one of the challenges is in mixing the fake and real time (used by select / poll) and generally syncing the threads. I run out of my QUIC time this week but will try to check again next week as this requires more research and understanding.