regex/src/compile.rs

238 lines
6.8 KiB
Rust

//!
//! Compiles the regex AST into a finite state machine
//!
//! The basic idea is to return the index and char transition to the new node, which will then
//! be put into the transitions of the current one
//!
//! Example regex: /a|b*c/
//!
//! Using {x} to reference items in the text below
//!
//! Has the AST
//! ```txt
//! Choice {1}
//! |-- Char(a)
//! |-- Seq {4}
//! | -- Repeat(Char(b)) {5}
//! | -- Char(c) {6}
//! ```
//! Should compile into the following state machine
//!
//! ```txt
//! {3} {2}
//! /--a--(())
//! ( START )
//! \--b---()--c--(())
//! /\
//! b_|
//! ```
//!
//! For that, we compile each node individually
//!
//! Compiling the choice node {2} will get us all nodes and transitions we need
//! Compiling the first char node will add a single node {2}.
//! This compilation will then return the index to this node and also the char it needs,
//! being the char that is contained in this char node. The choice node {2} will then
//! add a transition {3} to the start node it created.
//! Note that the choice node {2} does not now that it is the start node.
//!
//! The same is being done for the second child of the choice, although it's a bit more
//! complicated for that one.
//! First we compile the seq node {4}. This will directly lead to compiling it's two child nodes,
//! ({5}, {6}).
//! Compiling the repeat node {5} returns it's index and also the char that leads to it.
//! The char that leads to a repeat node is the one it repeats.
//! For the char node {6}, it's very similar to the char node below the choice node {1}.
//!
//! Another example: /u(w|o)!/
//!
//! AST:
//! ```txt
//! Seq
//! |-- Char(u)
//! |-- Choice
//! |-- Char(w)
//! |-- Char(o)
//! |-- Char(!)
//! ```
//!
//! ```txt
//!
//! /-w--()--\
//! ( START )--u--() |--!-(())
//! \-o--()--/
//!
//! ```
//!
//!
//! AST nodes will become transitions in the FSM
//! FSM nodes are the connections in the AST
//!
//! This architecture mostly seems to work out, with the only problem currently being allocating nodes
//! this appears to be something not every kind of regex part does.
//!
//! A char will allocate the node for its transition.
//! A seq won't to that, because the contents of the seq allocate everything, the seq is just a wrapper.
//! Now the question is: is seq unique and should be special cased, or can something like it exist?
//!
//! Does choice allocate a node? No, it does not, it only branches. So allocating seems like something
//! that some kinds do, but not all of them.
//!
//! So allocating is something that is not fundamental to the compilation, but handled by each node.
type NodeIndex = usize;
impl Compiler {
/// This function takes the node index of the previous node, constructs a new one as the target,
/// and then creates a transition from the previous to the new one, containing the condition
/// of the AST node it is compiling.
/// It returns
fn compile(&mut self, regex: &Regex, node_before: NodeIndex) -> NodeIndex {
match regex {
Regex::Char(char) => self.allocating(node_before, |_, _| TransitionType::Char(*char)),
Regex::Sequence(terms) => {
if let Some(first) = terms.first() {
let trans_to_first = self.compile(first, 0);
} else {
TransitionType::Always;
};
todo!()
}
Regex::Primitive(primitive) => self.allocating(node_before, |_, _| {
TransitionType::Primitive(match primitive {
parse::Primitive::Word => Primitive::Word,
parse::Primitive::Digit => Primitive::Digit,
})
}),
Regex::Choice(a, b) => {
todo!()
}
Regex::Repetition(_) => {
todo!()
}
Regex::Set(_) => {
todo!()
}
Regex::Range(_) => {
todo!()
}
}
}
fn allocating<F: FnOnce(&mut Node, NodeIndex) -> TransitionType>(
&mut self,
node_before: NodeIndex,
f: F,
) -> NodeIndex {
let next_node_slot = self.reserve_node_slot();
let mut next_node = Node::default();
let this_condition = f(&mut next_node, next_node_slot);
// fill the placeholder with the node we just created, forget the placeholder
let _ = std::mem::replace(self.nodes.get_mut(next_node_slot).unwrap(), next_node);
self.nodes
.get_mut(node_before)
.unwrap()
.transitions
.push(Transition {
target_node: next_node_slot,
condition: this_condition,
});
next_node_slot
}
}
use crate::parse;
use crate::parse::Regex;
use std::ops::Range;
#[derive(Debug, Clone, PartialEq, Eq)]
struct Transition {
target_node: usize,
condition: TransitionType,
}
#[derive(Debug, Clone, PartialEq, Eq)]
enum TransitionType {
Range(Range<char>),
Primitive(Primitive),
Char(char),
Always,
}
#[derive(Debug, Copy, Clone, PartialEq, Eq)]
enum Primitive {
Word,
Digit,
}
#[derive(Debug, Clone, Default, PartialEq, Eq)]
struct Node {
end: bool,
transitions: Vec<Transition>,
}
#[derive(Debug, Clone, PartialEq, Eq)]
struct RegexFsm {
nodes: Vec<Node>,
}
#[derive(Debug, Default)]
struct Compiler {
nodes: Vec<Node>,
}
/// Compiles the parsed Regex into a FSM
fn compile(regex: &Regex) -> RegexFsm {
let mut compiler = Compiler::default();
// reserve the start node
compiler.reserve_node_slot();
compiler.compile(regex, 0);
RegexFsm {
nodes: compiler.nodes,
}
}
impl Compiler {
/// Pushes a placeholder node into the internal buffer and returns it's index
fn reserve_node_slot(&mut self) -> NodeIndex {
self.nodes.push(Node::default());
self.nodes.len() - 1
}
}
#[cfg(test)]
mod test {
use crate::compile::{Node, RegexFsm, Transition, TransitionType};
use crate::parse::Regex;
///
/// regex: /🌈/
/// fsm: () --🌈-- (())
#[test]
fn single_char() {
let ast = Regex::Char('🌈');
let fsm = super::compile(&ast);
assert_eq!(
fsm,
RegexFsm {
nodes: vec![
Node {
end: false,
transitions: vec![Transition {
target_node: 1,
condition: TransitionType::Char('🌈')
}]
},
Node {
end: true,
transitions: vec![]
}
]
}
)
}
}