use std::rc::Rc; /// A location in the source text #[derive(Debug, Clone)] pub(crate) struct ParsedSpan { input: Rc, /// # Safety /// /// Must be a valid character boundary index into `input`. start: usize, /// # Safety /// /// Must be a valid character boundary index into `input`. end: usize, } #[derive(Debug, Clone)] pub(crate) enum Span { /// A location in the source text Parsed(ParsedSpan), /// Desugarings DuplicateRecordFieldsSugar, DottedFieldSugar, WithSugar, RecordPunSugar, /// For expressions obtained from decoding binary Decoded, /// For expressions constructed during normalization/typecheck Artificial, } impl ParsedSpan { pub(crate) fn to_input(&self) -> String { self.input.to_string() } /// Convert to a char range for consumption by annotate_snippets. /// This compensates for https://github.com/rust-lang/annotate-snippets-rs/issues/24 pub(crate) fn as_char_range(&self) -> (usize, usize) { ( char_idx_from_byte_idx(&self.input, self.start), char_idx_from_byte_idx(&self.input, self.end), ) } } impl Span { pub(crate) fn make(input: Rc, sp: pest::Span) -> Self { Span::Parsed(ParsedSpan { input, start: sp.start(), end: sp.end(), }) } /// Takes the union of the two spans, i.e. the range of input covered by the two spans plus any /// input between them. Assumes that the spans come from the same input. Fails if one of the /// spans does not point to an input location. pub(crate) fn union(&self, other: &Span) -> Self { use std::cmp::{max, min}; use Span::*; match (self, other) { (Parsed(x), Parsed(y)) if Rc::ptr_eq(&x.input, &y.input) => { Parsed(ParsedSpan { input: x.input.clone(), start: min(x.start, y.start), end: max(x.end, y.end), }) } _ => panic!( "Tried to union incompatible spans: {:?} and {:?}", self, other ), } } } /// Convert a byte idx into a string into a char idx for consumption by annotate_snippets. /// The byte idx must be at a char boundary. fn char_idx_from_byte_idx(input: &str, idx: usize) -> usize { use std::iter::once; input .char_indices() .map(|(byte_i, _)| byte_i) // We don't care about the char .chain(once(input.len())) // In case the idx points to the end of the string .enumerate() .find(|(_, byte_i)| *byte_i == idx) .map(|(char_i, _)| char_i) .unwrap() }