Nathaniel J. Smith

For the record, the reason I didn't do this in the first place is that I feel like `shield` should be a pretty obscure and rarely used feature. It's​ very...

I think there are two things making me uneasy here: * I don't feel like I have a very clear picture of what the goal here should be. Cancelling work...

Note: if we do promote graceful cancellation to a first-class concept, then I guess we'll want to rethink `restrict_keyboard_interrupt_to_checkpoints`. Maybe something like an argument that (a) makes it so KI...

> I am thinking along the lines of long running WebSockets where I am not too worried about losing a few messages but rather several hours of accumulated state. A...

> Can we have a "task that listens for control-C"? I thought this was usually caught by the outer nursery itself after all tasks had canceled. By default, `KeyboardInterrupt` gets...

I wonder if we should care that in principle someone might want to do some sort of graceful shutdown but doesn't care about the integrity of the stream they're reading...

This seems like a relevant paper for the whole "cancellation fragility" issue: [Kill-Safe Synchronization Abstractions](https://www.cs.utah.edu/plt/kill-safe/)

Was thinking about this again in the shower today. It seems like could maybe simplify the whole "soft shield" idea by making the default state be soft-shielded. That is, the...

The terminology here could use some work... graceful is on the tip of my tongue because I just stripped it out of `AsyncResource`, but we still have `trio.aclose_forcefully`, which is...

Another thing I just realized: for the minimal version of this idea where setting a grace period acts as (a) setting an overall deadline, and (b) setting a flag indicating...