badssl | Connect to broken ssl servers | TLS library

As the client, are all of these versions communicated to the server, and the server chooses the highest that it supports?

The client simply tells which versions are supported (TLS 1.3 supported_versions extension) or announces the best it can do (TLS 1.2 and lower). The server then simply picks the highest protocol version which is supported by both client and server.

If I denote a specific version SSLContext.getInstance("TLSv1.3",BCJSSE); and the server does not support that version is an exception thrown?

If there is no common protocol version supported by both client and server then the handshake will fail and an exception thrown.

I'm trying to determine why you would ever specify a version in your call, if the negotiation will automagically determine the best match.

This will usually only be done if there is a requirement to not support versions below a specific one, i.e. support only TLS 1.2 and higher. Since TLS 1.0 is considered too weak already in some situations, this can be a real-world requirement.

There are several Encrypted Handshake Message in the packet capture after the application data. This very likely means that the server itself does not request a client certificate by default but that the certificate is only requested for specific URL.

In this case first a TLS handshake is done without a CertificateRequest. Once the handshake is finished the client sends the HTTP request over the encrypted connection which is the Application Data in the packet capture. The server will determine that the requested URL needs a client certificate and initiate a renegotiation, i.e. another TLS handshake but this time with a CertificateRequest. But since the connection is already encrypted this renegotiation is only visible as Encrypted Handshake Message and the details cannot be seen without decrypting the traffic.

If I understood that correctly now, you don't want to change the default sort order, except for the parametrized tests.
Here is a slightly more complicated adapted version that shall do this (I tried to add enough comments to explain it):

You can change the order of the items in the pytest_collection_modifyitems hook. If you put this in your conftest.py:

The .NET X509Chain class has logic to find the missing certificates from ambient system state, which is kicking in here. The only time you'll see an incomplete chain from TLS is when both the server sent an incomplete chain and the system can't fill in the missing pieces.

SslLabs is reporting what happened specifically on the TLS connection. .NET's X509Chain considered the 2nd certificate, decided it made no sense in the chain, and threw it out, so you don't get to see it. There's no way to see the raw certificate data from the TLS messages via SslStream.

Try this:

It turns out that self-signed.badssl.com is expired. For some reason Python3 requests doesn't trust the chain that is passed as verify arg in case there is an expired cert in it.

Out of sheer desperation, I enabled Shared User Certificates in the capabilities section of my app's package manifest. It worked! Hopefully this helps others!