blog: doc usage of anyhow!

[JosiahParry]

Since extendr/extendr#973 returning anyhow::Result<> works without any modification.

I think this is a huge quality of life improvement. We can make a patch release with these features and document it?

For example:

use anyhow::bail;
use extendr_api::prelude::*;

#[extendr]
fn can_we_anyhow() -> anyhow::Result<()> {
    bail!("This is an error");
    Ok(())
}

lets us return a nice error

[yutannihilation]

Oh, cool. Congratulations!

[aquasync]

Hi, it looks like this was later reverted (in 5563f11). I'm confused as to what the recommendation is to have clean errors come through on the R side.

Ideally we have a rust function returning Result, which translates to an error in R. I've read through #278, and tried with/without result_condition feature. It seems the wrappers won't automatically call stop for you when using result_condition, so you then need to wrap every call-site with a condition check. Not sure if I'm missing something?

This change made sense to me - is there some way to regain the ability to have clean errors now?

Thanks!

[aquasync]

For the time being I'm using result_condition in combination with this (giant hack) in my package R code, so that I don't need to wrap every call myself:

.Call = function(...) {
    res = eval.parent(replace(match.call(), 1, list(base::.Call)))
    if (inherits(res, 'extendr_error')) stop(res$value, call.=F) else res
}

Might be useful for anyone else wanting a minimal effort stop-gap solution.

[JosiahParry]

"Hi, it looks like this was later reverted (in ). I'm confused as to what the recommendation is to have clean errors come through on the R side."

Where do you see this? It would be great if you could make an issue instead! 🙈

Edit: see extendr/extendr#1054

thanks!