I think CI failures must be unrelated to this PR, since I literally just added an export.

Yes, the time-manager was bumped to 2.0, and http2, the dependency of warp, uses it but that version hasn't been added to the stack.yml files used for the tests.
If you could add http2-5.3.9 to the extra-deps of all the stack*.ymls (except stack-nightly.yml), that'd be great 😅

@edsko Would you tell me why this exception should be exported? It is used only internally...

@Vlix I noticed this failure of CI and am trying to fix it. The fix should be done independent on this PR. So, @edsko does not have to tackle the CI issue.

CI has been fixed.

Would you tell me why this exception should be exported? It is used only internally...

Because it is critical that I am able to catch it.

Would you tell me why this exception should be exported? It is used only internally...

Specifically, in the http2 case, this exception is thrown to server handlers, which are running user-defined code (my code), and http2 sometimes kills those handlers when it considers them no longer required (when a client disappears), but I need to allow them to continue to run and finish what they were doing, so instead I catch the exception, mark the client as having disappeared in case the handler still tries to communicate, but allow the handler to continue to run to completion.

OK!
Rebased and merged.
Also, a new version has been released.

Thanks @kazu-yamamoto !