Replace Vec<Vec<T>> with a native Matrix representation

[hackaugusto]

Matrixes are currently represented with:

air-script/air-script-core/src/constant.rs

Line 41 in 0ed25aa

Matrix(Vec<Vec<u64>>),

The representation above has two issues:

• there is no guarantee the matrix size is correctly, i.e. the following can be created [[1], [2,3]]
• there is one allocation per row (probably not critical here, but always nice to reduce unnecessary allocations)

One sketch proposal is:

struct Matrix<T, const N: usize, const M: usize> {
    data: Vec<T>,
}

impl<T, const N: usize, const M: usize> Matrix<T, N, M> {
    fn new() -> Matrix<T, N, M> {
        let data = Vec::with_capacity(N * M);
        Matrix { data }
    }
}

Note: The above would be statically typed, another approach is to have a dynamic view of the data, like with pandas data frames.

[jan-ferdinand]

A crate worth considering is the imho excellent ndarray.

[hackaugusto]

Note: The validation of the matrix is currently done here:

air-script/ir/src/symbol_table/mod.rs

Lines 57 to 63 in 46586ea

	// check the number of elements in each row are same for a matrix
	if let ConstantValueExpr::Matrix(matrix) = &constant_type {
	let row_len = matrix[0].len();
	if matrix.iter().skip(1).any(|row| row.len() != row_len) {
	return Err(SemanticError::invalid_matrix_constant(&name));
	}
	}

Is the idea to air-script-core with parsed but not-yet-validated data?

Edit: I think it makes sense to have a separate type that performs validation in the constructor, it doesn't have to replace the structure I linked in the original issue, but having a type with these properties validated makes easier to audit (since one doesn't have to find were / if the validation is done).

[tohrnii]

Is the idea to air-script-core with parsed but not-yet-validated data?

We tried to keep validation out of the grammar in many cases in general to avoid making the grammar unnecessarily complex, but I agree with you that in this case it probably makes sense to do the validation in the constructor at the parser level.

[hackaugusto]

We tried to keep validation out of the grammar in many cases in general to avoid making the grammar unnecessarily complex

Thanks for the explanation, makes sense! How does the core crate fit into that?

[tohrnii]

How does the core crate fit into that?

The idea behind core crate was just to have commonly used structs/consts to be defined there (similar to miden-vm-core) so that it can be used as a dependency in other crates. For example, codegen doesn't depend on parser crate but imports some structs from core crate.