Document how to handle cancellation

[yutannihilation]

cf.

• https://github.com/krlmlr/cancel.test
• Can't cancel long-running queries duckdb/duckdb-r#514

[yutannihilation]

WRE suggests R_CheckUserInterrupt(), but it seems duckdb-r doesn't use it for some reason, which I don't figure out yet. But it seems it's just use csignal.

https://cran.rstudio.com/doc/manuals/r-devel/R-exts.html#Allowing-interrupts

[yutannihilation]

ctrlc crate looks handy.

https://rust-cli.github.io/book/in-depth/signals.html
https://github.com/Detegr/rust-ctrlc

[eitsupi]

This is very useful for me.

The current polars package implements the following, which I believe is not ideal.
eitsupi/neo-r-polars#23 (comment)
https://github.com/pola-rs/r-polars/blob/a7b064265f56ece53358955979f929b2cc055c63/src/rust/src/utils/extendr_concurrent.rs#L234-L250

[yutannihilation]

Oh, the current implementation looks very hacky...

I thought we can learn from Python Polars' implementation, but it seems they have their own difficulties...

pola-rs/polars#2687
https://pyo3.rs/v0.22.5/faq#ctrl-c-doesnt-do-anything-while-my-rust-code-is-executing

[yutannihilation]

I think this is the naive translation to Rust, but this doesn't work except for the first Ctrl+C...

https://github.com/yutannihilation/cancel.test.savvy/blob/main/src/rust/src/lib.rs

[yutannihilation]

Some more findings:

• It seems this is something related to the interaction with Positron. If I do the same thing on the bare R session on a terminal, Ctrl+C works as expected
• Calling eval_parse_text("Sys.sleep(0)")? on every loop is a workaround.
• However, considering krlmlr/cancel.test works without such a workaround, probably I'm missing some point
  • To be clear, krlmlr/cancel.test uses cpp11::message, which evaluates message call, but it works fine without it

[yutannihilation]

Calling eval_parse_text("Sys.sleep(0)")? on every loop is a workaround.

Other than Sys.sleep, message() and warning() works. However, most functions don't seem to work.