267 lines
8.2 KiB
Rust
267 lines
8.2 KiB
Rust
use std::ops::Range;
|
|
|
|
use gpui::{
|
|
App, Entity, Hsla, IntoElement, ParentElement, RenderOnce, SharedString, Styled, Window, div,
|
|
prelude::*,
|
|
};
|
|
|
|
use super::{Highlight, Input, TextInput};
|
|
use crate::theme::{ActiveTheme, ControlSize, TextSize};
|
|
|
|
/// What a span of source is.
|
|
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
|
|
pub(crate) enum Kind {
|
|
Comment,
|
|
Text,
|
|
Number,
|
|
Word,
|
|
Mark,
|
|
}
|
|
|
|
const KEYWORDS: &[&str] = &[
|
|
"as", "async", "await", "break", "case", "catch", "class", "const", "continue", "def", "do",
|
|
"else", "enum", "export", "fn", "for", "from", "func", "function", "if", "impl", "import",
|
|
"in", "let", "match", "mut", "new", "pub", "return", "static", "struct", "switch", "throw",
|
|
"try", "type", "use", "var", "while", "yield",
|
|
];
|
|
const CONSTANTS: &[&str] = &[
|
|
"false",
|
|
"nil",
|
|
"None",
|
|
"null",
|
|
"self",
|
|
"this",
|
|
"true",
|
|
"undefined",
|
|
];
|
|
|
|
/// Splits source into spans; spaces fall between them.
|
|
pub(crate) fn lex(source: &str) -> Vec<(Range<usize>, Kind)> {
|
|
let mut out = Vec::new();
|
|
let mut chars = source.char_indices().peekable();
|
|
while let Some((start, ch)) = chars.next() {
|
|
let kind = match ch {
|
|
'"' | '\'' | '`' => {
|
|
let mut escaped = false;
|
|
while let Some((_, next)) = chars.next_if(|(_, next)| *next != '\n') {
|
|
if escaped {
|
|
escaped = false;
|
|
} else if next == '\\' {
|
|
escaped = true;
|
|
} else if next == ch {
|
|
break;
|
|
}
|
|
}
|
|
Kind::Text
|
|
}
|
|
'/' if chars.peek().is_some_and(|(_, next)| *next == '/') => {
|
|
while chars.next_if(|(_, next)| *next != '\n').is_some() {}
|
|
Kind::Comment
|
|
}
|
|
'0'..='9' => {
|
|
while chars
|
|
.next_if(|(_, next)| next.is_ascii_alphanumeric() || matches!(next, '.' | '_'))
|
|
.is_some()
|
|
{}
|
|
Kind::Number
|
|
}
|
|
ch if ch.is_alphabetic() || ch == '_' || ch == '$' => {
|
|
while chars
|
|
.next_if(|(_, next)| next.is_alphanumeric() || matches!(next, '_' | '$'))
|
|
.is_some()
|
|
{}
|
|
Kind::Word
|
|
}
|
|
ch if ch.is_whitespace() => continue,
|
|
_ => Kind::Mark,
|
|
};
|
|
let end = chars.peek().map_or(source.len(), |(next, _)| *next);
|
|
out.push((start..end, kind));
|
|
}
|
|
out
|
|
}
|
|
|
|
fn painted(spans: impl Iterator<Item = (Range<usize>, Hsla)>) -> Vec<(Range<usize>, Highlight)> {
|
|
spans
|
|
.map(|(range, color)| {
|
|
let highlight = Highlight {
|
|
color,
|
|
background: None,
|
|
};
|
|
(range, highlight)
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Colors for one line of code in most languages, from the syntax palette.
|
|
pub fn code_highlights(source: &str, cx: &App) -> Vec<(Range<usize>, Highlight)> {
|
|
let syntax = &cx.theme().colors.syntax;
|
|
let spans = lex(source);
|
|
let next_is = |ix: usize, mark: &str| {
|
|
spans
|
|
.get(ix + 1)
|
|
.is_some_and(|(range, _)| &source[range.clone()] == mark)
|
|
};
|
|
painted(spans.iter().enumerate().map(|(ix, (range, kind))| {
|
|
let text = &source[range.clone()];
|
|
let color = match kind {
|
|
Kind::Comment => syntax.comment,
|
|
Kind::Text => syntax.string,
|
|
Kind::Number => syntax.number,
|
|
Kind::Word if KEYWORDS.contains(&text) => syntax.keyword,
|
|
Kind::Word if CONSTANTS.contains(&text) => syntax.constant,
|
|
Kind::Word if next_is(ix, "(") => syntax.function,
|
|
Kind::Word if text.starts_with(char::is_uppercase) => syntax.type_name,
|
|
Kind::Word => syntax.variable,
|
|
Kind::Mark if "()[]{},.;".contains(text) => syntax.punctuation,
|
|
Kind::Mark => syntax.operator,
|
|
};
|
|
(range.clone(), color)
|
|
}))
|
|
}
|
|
|
|
/// Colors for JSON: keys, strings, numbers and literals.
|
|
pub fn json_highlights(source: &str, cx: &App) -> Vec<(Range<usize>, Highlight)> {
|
|
let syntax = &cx.theme().colors.syntax;
|
|
let spans = lex(source);
|
|
let keyed = |ix: usize| {
|
|
spans
|
|
.get(ix + 1)
|
|
.is_some_and(|(range, _)| &source[range.clone()] == ":")
|
|
};
|
|
painted(spans.iter().enumerate().filter_map(|(ix, (range, kind))| {
|
|
let color = match kind {
|
|
Kind::Text if keyed(ix) => syntax.property,
|
|
Kind::Text => syntax.string,
|
|
Kind::Number => syntax.number,
|
|
Kind::Word if matches!(&source[range.clone()], "true" | "false" | "null") => {
|
|
syntax.constant
|
|
}
|
|
Kind::Mark => syntax.punctuation,
|
|
Kind::Word | Kind::Comment => return None,
|
|
};
|
|
Some((range.clone(), color))
|
|
}))
|
|
}
|
|
|
|
/// A one-line code field in the mono face. Give its state `code_highlights`.
|
|
#[derive(IntoElement)]
|
|
pub struct CodeInput {
|
|
state: Entity<TextInput>,
|
|
size: ControlSize,
|
|
}
|
|
|
|
impl CodeInput {
|
|
pub fn new(state: &Entity<TextInput>) -> Self {
|
|
Self {
|
|
state: state.clone(),
|
|
size: ControlSize::default(),
|
|
}
|
|
}
|
|
|
|
pub fn size(mut self, size: ControlSize) -> Self {
|
|
self.size = size;
|
|
self
|
|
}
|
|
}
|
|
|
|
impl RenderOnce for CodeInput {
|
|
fn render(self, _: &mut Window, cx: &mut App) -> impl IntoElement {
|
|
assert!(
|
|
!self.state.read(cx).is_multi_line(),
|
|
"a code input is one line; JsonInput takes several"
|
|
);
|
|
div()
|
|
.w_full()
|
|
.font_family(cx.theme().mono_family.clone())
|
|
.child(Input::new(&self.state).size(self.size))
|
|
}
|
|
}
|
|
|
|
/// A JSON field, checked as you type. Give its state `json_highlights`.
|
|
#[derive(IntoElement)]
|
|
pub struct JsonInput {
|
|
state: Entity<TextInput>,
|
|
}
|
|
|
|
impl JsonInput {
|
|
pub fn new(state: &Entity<TextInput>) -> Self {
|
|
Self {
|
|
state: state.clone(),
|
|
}
|
|
}
|
|
}
|
|
|
|
impl RenderOnce for JsonInput {
|
|
fn render(self, _: &mut Window, cx: &mut App) -> impl IntoElement {
|
|
let source = self.state.read(cx).text().to_string();
|
|
let problem = (!source.trim().is_empty())
|
|
.then(|| serde_json::from_str::<serde_json::Value>(&source).err())
|
|
.flatten()
|
|
.map(|error| SharedString::from(error.to_string()));
|
|
let theme = cx.theme();
|
|
div()
|
|
.flex()
|
|
.flex_col()
|
|
.gap_1()
|
|
.w_full()
|
|
.child(
|
|
div()
|
|
.font_family(theme.mono_family.clone())
|
|
.child(Input::new(&self.state).invalid(problem.is_some())),
|
|
)
|
|
.when_some(problem, |field, problem| {
|
|
field.child(
|
|
div()
|
|
.text_size(theme.text_size(TextSize::Xs))
|
|
.text_color(theme.colors.danger)
|
|
.child(problem),
|
|
)
|
|
})
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::{Kind, lex};
|
|
|
|
fn kinds(source: &str) -> Vec<(&str, Kind)> {
|
|
lex(source)
|
|
.into_iter()
|
|
.map(|(range, kind)| (&source[range], kind))
|
|
.collect()
|
|
}
|
|
|
|
#[test]
|
|
fn lexing_splits_strings_numbers_words_and_comments() {
|
|
assert_eq!(
|
|
kinds(r#"let s = "a \"b\""; // done"#),
|
|
[
|
|
("let", Kind::Word),
|
|
("s", Kind::Word),
|
|
("=", Kind::Mark),
|
|
(r#""a \"b\"""#, Kind::Text),
|
|
(";", Kind::Mark),
|
|
("// done", Kind::Comment),
|
|
]
|
|
);
|
|
assert_eq!(
|
|
kinds("f(0x1F, 2.5)"),
|
|
[
|
|
("f", Kind::Word),
|
|
("(", Kind::Mark),
|
|
("0x1F", Kind::Number),
|
|
(",", Kind::Mark),
|
|
("2.5", Kind::Number),
|
|
(")", Kind::Mark),
|
|
]
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn an_open_string_stops_at_the_line_end() {
|
|
assert_eq!(kinds("'open"), [("'open", Kind::Text)]);
|
|
assert_eq!(kinds("\"a\n1"), [("\"a", Kind::Text), ("1", Kind::Number)]);
|
|
}
|
|
}
|