Steven Fackler

Is this in reference to a connection implementation that will be in a bad state during a panic, or is this more of a general principle of the thing?

The fact that I'm not aware of any instance in which this has been a problem in the last several years guides my instincts to some extent 😃 . That...

I don't see how that's really related? Other kinds of "panic safety" like mutex poisoning has been around for quite a long time.

Can you not use RAII for your transaction management?

I mean rolling back the connection on Drop if the transaction hasn't been committed: https://github.com/sfackler/rust-postgres/blob/master/tokio-postgres/src/transaction.rs#L30

Why? parking_lot also avoids the poisoning behavior.

I don't know if comparing the string description of the underlying error is the best idea.

Why do you need to compare your wrapped error for equality? There are a *lot* of error types in the wild that cannot do that (e.g. `io::Error`, `Box`, ...).

What kind of hardware is your CI running? I cannot reproduce this failure on an M1 macbook (either in macOS or Linux via Docker).

How does something like Hikari handle validity checks?