Hacker News Reader

It does, but you can only run synchronous code on drop. This is what "async drop" is supposed to handle — things like issuing ROLLBACK statements to the database on cancellation.

It also wouldn't help when you have no valid state to restore to, as in the mutex example in the post.

View on HN

tokio-postgres handles this by just dispatching the "ROLLBACK" command in impl Drop and ignoring the response. https://github.com/rust-postgres/rust-postgres/blob/a7a49a90...

Is this not enough? What could go wrong? If the network connection dies or the task is cancelled, I'm assuming the database server cleans up the connection state and does a rollback automatically.

And adding async Drop will probably add a whole new set of footguns.

> What could go wrong?

LoL, an insane amount of things. TCP connections are an illusion of safely, for the purpose of database commits use UDP packets as a model instead, it'll be much closer to reality.

> an insane amount of things

List a couple

> TCP connections are an illusion of safely

Why?

Reply on news.ycombinator.com