More try_from conversions

[andy-thomason]

This PR covers i32, f64, Bool and u8 types in vectors and slices.

        assert_eq!(<Vec::<i32>>::try_from(Robj::from(1)), Ok(vec![1]));
        assert_eq!(<Vec::<f64>>::try_from(Robj::from(1.)), Ok(vec![1.]));
        assert_eq!(<Vec::<Bool>>::try_from(Robj::from(TRUE)), Ok(vec![TRUE]));
        assert_eq!(<Vec::<u8>>::try_from(Robj::from(0_u8)), Ok(vec![0_u8]));

        assert_eq!(<&[i32]>::try_from(Robj::from(1)), Ok(&[1][..]));
        assert_eq!(<&[f64]>::try_from(Robj::from(1.)), Ok(&[1.][..]));
        assert_eq!(<&[Bool]>::try_from(Robj::from(TRUE)), Ok(&[TRUE][..]));
        assert_eq!(<&[u8]>::try_from(Robj::from(0_u8)), Ok(&[0_u8][..]));

[clauswilke]

Looks good overall! I have a few comments and questions, though.

[clauswilke]

The thing that I don't fully understand yet: Shouldn't the contents of the vectors/slices also be converted with try_from()? It looks to me like we're applying different rules for scalars and vectors/slices right now, e.g. auto-conversion of floats to integers and back.

[clauswilke]

Same as above. Shouldn't this succeed?

[andy-thomason]

Ditto.

[andy-thomason]

The slices are a bit of an exception as they give immutable reference access to the interiors of the R objects.

You can still check for NA using value.is_na() but checking all the values of a gigabyte-long vector may be undesirable.

[andy-thomason]

We should consider having mutable slices also &mut [i32] etc.

This would allow the modification of R objects in-place.

[clauswilke]

Ok, I get the argument for the slices, but not for vectors. There we're making a copy of the data, correct? So then we should use the same conversion and rules that we're using for scalars (convert 1.0 to 1 etc.)

[yutannihilation]

The slices are a bit of an exception as they give immutable reference access to the interiors of the R objects.

So, this is also the reason why no other int/float variants (e.g. i64 and f32) are supported in slice's case, right? I'll document the rule on #244.

[clauswilke]

I've decided to approve, as this PR moves the project forward.

However, the issue of numerical vectors is not fully resolved, as R doesn't do a good job of distinguishing between integer and floating point values. We need some sort of option that allows a Rust function to accept either integer or float vectors as input and use them as either floats or integers, with auto-conversion as appropriate.

[yutannihilation]

We need some sort of option that allows a Rust function to accept either integer or float vectors as input and use them as either floats or integers, with auto-conversion as appropriate.

I agree. I think we might need to introduce some wrapper like Nullable. Or, if it's not necessarily on Rust's side, call as.xxx() function in the wrapper R function?