vctrsrs

The goal of vctrsrs is to allow extendr developers to return a single struct to R that can be interacted with like a normal vector.

This is based on extendr_usize_vector sample repo.

Example

Here is an example of a Vec<Option<usize>>

library(vctrsrs)

x <- new_usize(sample(1e5, 10))

length(x)

## [1] 10

x[1:10]

## <Vctr[10]> 
##  [1] "88786" "81486" "11988" "12025" "5214"  "36683" "39256" "30246" "32422"
## [10] "66222"

x[c(NA, -1, 5)]

## <Vctr[3]> 
## [1] "NA"   "NA"   "5214"

data.frame(id = 1:10, x = x)

##    id     x
## 1   1 88786
## 2   2 81486
## 3   3 11988
## 4   4 12025
## 5   5  5214
## 6   6 36683
## 7   7 39256
## 8   8 30246
## 9   9 32422
## 10 10 66222

how it works

In the Rust source a trait and a struct are implemented. The trait Rvctr looks like so:

pub trait Rvctr {
    fn show(&self) -> Strings;
    fn length(&self) -> Rint ;
    fn subset(self, idx: Integers) -> Self;
    fn extend(self, y: Self) -> Self;
}

These methods are all that are needed to make a single pointed feel like a vector by providing users the ability to use length(), c(), [, and [[.

The struct Vctr is intended to be a general purpose wrapper struct for any custom struct. It is a tuple struct that accepts and object that has implemented the Rvctr trait. It looks like so:

pub struct Vctr<T: Rvctr>(pub T);

You then generate an extendr R object by defining the impl with the #[extendr] macro. In this impl you have to also specify the methods for the trait which should be autopopulated with a custom macro like #[vctr] which isn’t yet made.

// this could be created by a macro...
#[extendr]
impl Vctr<VecUsize> {
    fn length(&self) -> Rint {
        self.0.length()
    }

    fn show(&self) -> Strings {
        self.0.show()
    }

    fn subset(&mut self, idx: Integers) -> Self {
        let x = self.0.clone();
        let res = x.subset(idx);
        Vctr(res)
    }
}

Problem: pointers aren’t vectors

This implementation can be put into a data frame. But it can’t be put into a tibble. That’s because tibble does all of its validation with vctrs. And vctrs only works with atomic and list vectors. So even though all of this witch craft has been implemented here. It cannot be compatible with the tidyverse unless it can pass vctrs::vec_is().

data.frame(id = 1:10, x = x)

##    id     x
## 1   1 88786
## 2   2 81486
## 3   3 11988
## 4   4 12025
## 5   5  5214
## 6   6 36683
## 7   7 39256
## 8   8 30246
## 9   9 32422
## 10 10 66222

tibble::tibble(id = 1:10, x = x)

## Error:
## ! All columns in a tibble must be vectors.
## ✖ Column `x` is a `Vctr` object.