aboutsummaryrefslogtreecommitdiffstats
path: root/src/construct/attention.rs
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/construct/attention.rs66
1 files changed, 39 insertions, 27 deletions
diff --git a/src/construct/attention.rs b/src/construct/attention.rs
index 65c2f6f..b042645 100644
--- a/src/construct/attention.rs
+++ b/src/construct/attention.rs
@@ -110,23 +110,23 @@ enum MarkerKind {
}
impl MarkerKind {
- /// Turn the kind into a [char].
- fn as_char(&self) -> char {
+ /// Turn the kind into a byte ([u8]).
+ fn as_byte(&self) -> u8 {
match self {
- MarkerKind::Asterisk => '*',
- MarkerKind::Underscore => '_',
+ MarkerKind::Asterisk => b'*',
+ MarkerKind::Underscore => b'_',
}
}
- /// Turn [char] into a kind.
+ /// Turn a byte ([u8]) into a kind.
///
/// ## Panics
///
- /// Panics if `char` is not `*` or `_`.
- fn from_char(char: char) -> MarkerKind {
- match char {
- '*' => MarkerKind::Asterisk,
- '_' => MarkerKind::Underscore,
- _ => unreachable!("invalid char"),
+ /// Panics if `byte` is not `*` or `_`.
+ fn from_byte(byte: u8) -> MarkerKind {
+ match byte {
+ b'*' => MarkerKind::Asterisk,
+ b'_' => MarkerKind::Underscore,
+ _ => unreachable!("invalid byte"),
}
}
}
@@ -160,9 +160,9 @@ struct Sequence {
/// ```
pub fn start(tokenizer: &mut Tokenizer) -> State {
match tokenizer.current {
- Some(char) if tokenizer.parse_state.constructs.attention && matches!(char, '*' | '_') => {
+ Some(byte) if tokenizer.parse_state.constructs.attention && matches!(byte, b'*' | b'_') => {
tokenizer.enter(Token::AttentionSequence);
- inside(tokenizer, MarkerKind::from_char(char))
+ inside(tokenizer, MarkerKind::from_byte(byte))
}
_ => State::Nok,
}
@@ -175,7 +175,7 @@ pub fn start(tokenizer: &mut Tokenizer) -> State {
/// ^^
/// ```
fn inside(tokenizer: &mut Tokenizer, marker: MarkerKind) -> State {
- if tokenizer.current == Some(marker.as_char()) {
+ if tokenizer.current == Some(marker.as_byte()) {
tokenizer.consume();
State::Fn(Box::new(move |t| inside(t, marker)))
} else {
@@ -188,7 +188,6 @@ fn inside(tokenizer: &mut Tokenizer, marker: MarkerKind) -> State {
/// Resolve attention sequences.
#[allow(clippy::too_many_lines)]
fn resolve_attention(tokenizer: &mut Tokenizer) {
- let chars = &tokenizer.parse_state.chars;
let mut start = 0;
let mut balance = 0;
let mut sequences = vec![];
@@ -203,21 +202,34 @@ fn resolve_attention(tokenizer: &mut Tokenizer) {
if enter.token_type == Token::AttentionSequence {
let end = start + 1;
let exit = &tokenizer.events[end];
- let marker =
- MarkerKind::from_char(Slice::from_point(chars, &enter.point).head().unwrap());
+
+ let before_end = enter.point.index;
+ let before_start = if before_end < 4 { 0 } else { before_end - 4 };
+ let string_before =
+ String::from_utf8_lossy(&tokenizer.parse_state.bytes[before_start..before_end]);
+ let char_before = string_before.chars().last();
+
+ let after_start = exit.point.index;
+ let after_end = if after_start + 4 > tokenizer.parse_state.bytes.len() {
+ tokenizer.parse_state.bytes.len()
+ } else {
+ after_start + 4
+ };
+ let string_after =
+ String::from_utf8_lossy(&tokenizer.parse_state.bytes[after_start..after_end]);
+ let char_after = string_after.chars().next();
+
+ let marker = MarkerKind::from_byte(
+ Slice::from_point(tokenizer.parse_state.bytes, &enter.point)
+ .head()
+ .unwrap(),
+ );
let before = classify_character(if enter.point.index > 0 {
- Slice::from_point(
- chars,
- &Point {
- index: enter.point.index - 1,
- ..enter.point
- },
- )
- .tail()
+ char_before
} else {
None
});
- let after = classify_character(Slice::from_point(chars, &exit.point).tail());
+ let after = classify_character(char_after);
let open = after == GroupKind::Other
|| (after == GroupKind::Punctuation && before != GroupKind::Other);
// To do: GFM strikethrough?
@@ -490,7 +502,7 @@ fn resolve_attention(tokenizer: &mut Tokenizer) {
/// * [`micromark-util-classify-character` in `micromark`](https://github.com/micromark/micromark/blob/main/packages/micromark-util-classify-character/dev/index.js)
fn classify_character(char: Option<char>) -> GroupKind {
match char {
- // Custom characters.
+ // EOF.
None => GroupKind::Whitespace,
// Unicode whitespace.
Some(char) if char.is_whitespace() => GroupKind::Whitespace,