Releasing v2.2.0

[kazu-yamamoto]

Relaxing boundary on Hackage

• crypton-connection
• http3

Fixing and releasing

• tls
• warp-tls
• http2-tls

Cc: @edsko @FinleyMcIlwaine

[kazu-yamamoto]

bye is used in connectionClose of crypton-connection which is typically used in bracket, sigh.

[kazu-yamamoto]

bye could be implemented without timeout which uses the asynchronous exception.
We need two parts:

(1) TimeManager of GHC: this is easy
(2) Non-blocking recv: this is possible with the normal blocking recv if we can check its readability. We have such a feature for Socket. But I don't know such a feature for Handle.

We need to consider dropping the support for Handle in tls.

[kazu-yamamoto]

It appeared that timeout can be used in bracket. haskell/network#589

• The version should be reverted into v2.1.
• The doc of bye should be relaxed.
• unliftio should be removed.

[edsko]

Just for my understanding, what do you mean by this?

The doc of bye should be relaxed.

[kazu-yamamoto]

I wrote that bye MUST NOT be used with bracket but now the usage is allowed as it used to be. (6eb4e53)

[edsko]

Hmm, but I think there is some harm at least in using it in a bracket pattern. If the connection is broken, resulting in an exception in the body, the call to bye will then try to initiate the TLS shutdown sequence, resulting in a different exception, obscuring the original, unless we swallow exceptions in bye (but that would then mean that if the TLS shutdown sequence itself fails, we'd not see the exception, which is potentially problematic also).

But fair enough, I don't feel terribly strongly here :) The recommended usage in the docs is okay with me.