Note on Recursion

Consider again the example shown in the beginning of the book:

E  => T E'
E' => + T E' | ε
T  => F T'
T' => * F T' | ε
F  => ( E ) | id

Note that this grammar involves recursions. For example, the productions of E' and T' involve themselves; and production of E involves F indirectly, which in turn involves E.

There are 2 things to note when implement a grammar that has recursion:

The data structure must have a finite size. This can be easily achieved by using a Box.
There must NOT be recursion when computing if trait requirements are satisfied.

We will go through an example to illustrate both points. Suppose we are implementing E' as:

#![allow(unused)]
fn main() {
// E' => + T E' | ε
#[derive_syntax]
struct Eprime(tp::Option<(OpAdd, T, Eprime)>);
}

Note we are using tp::Option to represent the ε case. This is a utility type used to represent optional values. Utility types will be covered in the next chapter.

The first error you will get is:

recursive type `Eprime` has infinite size

This is because Eprime contains Eprime itself, creating an infinite recursion when calculating the size. To fix this, we can box the recursive part:

#![allow(unused)]
fn main() {
#[derive_syntax]
struct Eprime(tp::Option<(OpAdd, T, Box<Eprime>)>);
}

Now we will see the second error:

Overflow evaluating the requirement `Eprime: Produce`

This error message is a little cryptic. One might think it's related to having Eprime referenced in itself. However, the root cause is that derive_syntax infers the lexicon type by looking at the first fields of structs and first variants of enums. We can walk through how Rust is evaluating this:

The lexicon type of Eprime is the same as the lexicon type of tp::Option<(OpAdd, T, Box<Eprime>)>.
The lexicon type of tp::Option<(OpAdd, T, Box<Eprime>)> is the same as that of (OpAdd, T, Box<Eprime>).
The lexicon type of (OpAdd, T, Box<Eprime>) is the same as that of OpAdd, T, and Box<Eprime>, and they must be the same (according to the trait implementation for tuples)
The lexicon type of OpAdd is TokenType.
The lexicon type of T is TokenType, which is the same as that of OpAdd.
The lexicon type of Box<Eprime> is the same as that of Eprime.
We are already calculating the lexicon type of Eprime, resulting in overflow

To fix this, the example in the beginning of the book uses a separate struct Eplus to avoid the loop:

#![allow(unused)]
fn main() {
#[derive_syntax]
struct Eprime(tp::Option<Eplus>);

#[derive_syntax]
struct Eplus {
    op: OpAdd,
    _t: T,
    rest: Box<Eprime>,
}
}

In this case, the lexicon type of Eplus does not depend on Eprime, because derive_syntax generates an implementation for Eplus instead of relying on blanket trait implementation as in the tuple case.

Another place where this error can appear is in enums. In the following example, Paren must not be the first variant of the enum.

#![allow(unused)]
fn main() {
#[derive_syntax]
enum F {
    Ident(Ident),
    Paren((ParenOpen, Box<E>, ParenClose)),
}
}

Remove Recursion with Utility Types

In the first example above, the grammar requires recursion in E' to implement a list-like structure (terms separated by +). Teleparse provides utility types like tp::Split to simplify the implementation and provide out-of-the-box support for panic recovery. Utility types are covered in Chapter 4.

The Teleparse Book

Note on Recursion

Remove Recursion with Utility Types