mirror of
https://github.com/denoland/deno.git
synced 2024-11-22 15:06:54 -05:00
d5634164cb
Closes https://github.com/denoland/deno/issues/2699 Closes https://github.com/denoland/deno/issues/2347 Uses unstable rustfmt features. Since dprint invokes `rustfmt` we do not need to switch the cargo toolchain to nightly. Do we care about formatting stability of our codebase across Rust versions? (I don't)
355 lines
8.9 KiB
Rust
355 lines
8.9 KiB
Rust
// Copyright 2018-2023 the Deno authors. All rights reserved. MIT license.
|
|
|
|
// The logic of this module is heavily influenced by
|
|
// https://github.com/microsoft/vscode/blob/main/extensions/typescript-language-features/src/languageFeatures/semanticTokens.ts
|
|
// and https://github.com/microsoft/vscode/blob/main/src/vs/workbench/api/common/extHostTypes.ts
|
|
// for the SemanticTokensBuilder implementation.
|
|
|
|
use std::ops::Index;
|
|
use std::ops::IndexMut;
|
|
use tower_lsp::lsp_types::SemanticToken;
|
|
use tower_lsp::lsp_types::SemanticTokenModifier;
|
|
use tower_lsp::lsp_types::SemanticTokenType;
|
|
use tower_lsp::lsp_types::SemanticTokens;
|
|
use tower_lsp::lsp_types::SemanticTokensLegend;
|
|
|
|
pub const MODIFIER_MASK: u32 = 255;
|
|
pub const TYPE_OFFSET: u32 = 8;
|
|
|
|
enum TokenType {
|
|
Class = 0,
|
|
Enum = 1,
|
|
Interface = 2,
|
|
Namespace = 3,
|
|
TypeParameter = 4,
|
|
Type = 5,
|
|
Parameter = 6,
|
|
Variable = 7,
|
|
EnumMember = 8,
|
|
Property = 9,
|
|
Function = 10,
|
|
Method = 11,
|
|
}
|
|
|
|
impl<T> Index<TokenType> for Vec<T> {
|
|
type Output = T;
|
|
fn index(&self, idx: TokenType) -> &T {
|
|
&self[idx as usize]
|
|
}
|
|
}
|
|
|
|
impl<T> IndexMut<TokenType> for Vec<T> {
|
|
fn index_mut(&mut self, idx: TokenType) -> &mut T {
|
|
&mut self[idx as usize]
|
|
}
|
|
}
|
|
|
|
enum TokenModifier {
|
|
Declaration = 0,
|
|
Static = 1,
|
|
Async = 2,
|
|
Readonly = 3,
|
|
DefaultLibrary = 4,
|
|
Local = 5,
|
|
}
|
|
|
|
impl<T> Index<TokenModifier> for Vec<T> {
|
|
type Output = T;
|
|
fn index(&self, idx: TokenModifier) -> &T {
|
|
&self[idx as usize]
|
|
}
|
|
}
|
|
|
|
impl<T> IndexMut<TokenModifier> for Vec<T> {
|
|
fn index_mut(&mut self, idx: TokenModifier) -> &mut T {
|
|
&mut self[idx as usize]
|
|
}
|
|
}
|
|
|
|
pub fn get_legend() -> SemanticTokensLegend {
|
|
let mut token_types = vec![SemanticTokenType::from(""); 12];
|
|
token_types[TokenType::Class] = "class".into();
|
|
token_types[TokenType::Enum] = "enum".into();
|
|
token_types[TokenType::Interface] = "interface".into();
|
|
token_types[TokenType::Namespace] = "namespace".into();
|
|
token_types[TokenType::TypeParameter] = "typeParameter".into();
|
|
token_types[TokenType::Type] = "type".into();
|
|
token_types[TokenType::Parameter] = "parameter".into();
|
|
token_types[TokenType::Variable] = "variable".into();
|
|
token_types[TokenType::EnumMember] = "enumMember".into();
|
|
token_types[TokenType::Property] = "property".into();
|
|
token_types[TokenType::Function] = "function".into();
|
|
token_types[TokenType::Method] = "method".into();
|
|
|
|
let mut token_modifiers = vec![SemanticTokenModifier::from(""); 6];
|
|
token_modifiers[TokenModifier::Async] = "async".into();
|
|
token_modifiers[TokenModifier::Declaration] = "declaration".into();
|
|
token_modifiers[TokenModifier::Readonly] = "readonly".into();
|
|
token_modifiers[TokenModifier::Static] = "static".into();
|
|
token_modifiers[TokenModifier::Local] = "local".into();
|
|
token_modifiers[TokenModifier::DefaultLibrary] = "defaultLibrary".into();
|
|
|
|
SemanticTokensLegend {
|
|
token_types,
|
|
token_modifiers,
|
|
}
|
|
}
|
|
|
|
pub struct SemanticTokensBuilder {
|
|
prev_line: u32,
|
|
prev_char: u32,
|
|
data_is_sorted_and_delta_encoded: bool,
|
|
data: Vec<u32>,
|
|
}
|
|
|
|
impl SemanticTokensBuilder {
|
|
pub fn new() -> Self {
|
|
Self {
|
|
prev_line: 0,
|
|
prev_char: 0,
|
|
data_is_sorted_and_delta_encoded: true,
|
|
data: Vec::new(),
|
|
}
|
|
}
|
|
|
|
pub fn push(
|
|
&mut self,
|
|
line: u32,
|
|
char: u32,
|
|
length: u32,
|
|
token_type: u32,
|
|
token_modifiers: u32,
|
|
) {
|
|
if self.data_is_sorted_and_delta_encoded
|
|
&& (line < self.prev_line
|
|
|| (line == self.prev_line && char < self.prev_char))
|
|
{
|
|
// push calls were ordered and are no longer ordered
|
|
self.data_is_sorted_and_delta_encoded = false;
|
|
|
|
// Remove delta encoding from data
|
|
let token_count = self.data.len() / 5;
|
|
let mut prev_line = 0;
|
|
let mut prev_char = 0;
|
|
for i in 0..token_count {
|
|
let mut line = self.data[5 * i];
|
|
let mut char = self.data[5 * i + 1];
|
|
|
|
if line == 0 {
|
|
// on the same line as previous token
|
|
line = prev_line;
|
|
char += prev_char;
|
|
} else {
|
|
// on a different line than previous token
|
|
line += prev_line;
|
|
}
|
|
|
|
self.data[5 * i] = line;
|
|
self.data[5 * i + 1] = char;
|
|
|
|
prev_line = line;
|
|
prev_char = char;
|
|
}
|
|
}
|
|
|
|
let mut push_line = line;
|
|
let mut push_char = char;
|
|
if self.data_is_sorted_and_delta_encoded && !self.data.is_empty() {
|
|
push_line -= self.prev_line;
|
|
if push_line == 0 {
|
|
push_char -= self.prev_char;
|
|
}
|
|
}
|
|
|
|
self.data.reserve(5);
|
|
self.data.push(push_line);
|
|
self.data.push(push_char);
|
|
self.data.push(length);
|
|
self.data.push(token_type);
|
|
self.data.push(token_modifiers);
|
|
|
|
self.prev_line = line;
|
|
self.prev_char = char;
|
|
}
|
|
|
|
fn data_to_semantic_token_vec(
|
|
data: &[u32],
|
|
data_is_sorted_and_delta_encoded: bool,
|
|
) -> Vec<SemanticToken> {
|
|
let token_count = data.len() / 5;
|
|
let mut result: Vec<SemanticToken> = Vec::with_capacity(token_count);
|
|
if data_is_sorted_and_delta_encoded {
|
|
for i in 0..token_count {
|
|
let src_offset = 5 * i;
|
|
result.push(SemanticToken {
|
|
delta_line: data[src_offset],
|
|
delta_start: data[src_offset + 1],
|
|
length: data[src_offset + 2],
|
|
token_type: data[src_offset + 3],
|
|
token_modifiers_bitset: data[src_offset + 4],
|
|
});
|
|
}
|
|
return result;
|
|
}
|
|
|
|
let mut pos: Vec<usize> = (0..token_count).collect();
|
|
pos.sort_by(|a, b| {
|
|
let a_line = data[5 * a];
|
|
let b_line = data[5 * b];
|
|
if a_line == b_line {
|
|
let a_char = data[5 * a + 1];
|
|
let b_char = data[5 * b + 1];
|
|
return a_char.cmp(&b_char);
|
|
}
|
|
a_line.cmp(&b_line)
|
|
});
|
|
|
|
let mut prev_line = 0;
|
|
let mut prev_char = 0;
|
|
for i in pos.iter() {
|
|
let src_offset = 5 * i;
|
|
let line = data[src_offset];
|
|
let char = data[src_offset + 1];
|
|
let length = data[src_offset + 2];
|
|
let token_type = data[src_offset + 3];
|
|
let token_modifiers_bitset = data[src_offset + 4];
|
|
|
|
let delta_line = line - prev_line;
|
|
let delta_start = if delta_line == 0 {
|
|
char - prev_char
|
|
} else {
|
|
char
|
|
};
|
|
|
|
result.push(SemanticToken {
|
|
delta_line,
|
|
delta_start,
|
|
length,
|
|
token_type,
|
|
token_modifiers_bitset,
|
|
});
|
|
|
|
prev_line = line;
|
|
prev_char = char;
|
|
}
|
|
|
|
result
|
|
}
|
|
|
|
pub fn build(&self, result_id: Option<String>) -> SemanticTokens {
|
|
SemanticTokens {
|
|
result_id,
|
|
data: SemanticTokensBuilder::data_to_semantic_token_vec(
|
|
&self.data,
|
|
self.data_is_sorted_and_delta_encoded,
|
|
),
|
|
}
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::*;
|
|
|
|
#[test]
|
|
fn test_semantic_tokens_builder_simple() {
|
|
let mut builder = SemanticTokensBuilder::new();
|
|
builder.push(1, 0, 5, 1, 1);
|
|
builder.push(1, 10, 4, 2, 2);
|
|
builder.push(2, 2, 3, 2, 2);
|
|
assert_eq!(
|
|
builder.build(None).data,
|
|
vec![
|
|
SemanticToken {
|
|
delta_line: 1,
|
|
delta_start: 0,
|
|
length: 5,
|
|
token_type: 1,
|
|
token_modifiers_bitset: 1
|
|
},
|
|
SemanticToken {
|
|
delta_line: 0,
|
|
delta_start: 10,
|
|
length: 4,
|
|
token_type: 2,
|
|
token_modifiers_bitset: 2
|
|
},
|
|
SemanticToken {
|
|
delta_line: 1,
|
|
delta_start: 2,
|
|
length: 3,
|
|
token_type: 2,
|
|
token_modifiers_bitset: 2
|
|
}
|
|
]
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn test_semantic_tokens_builder_out_of_order_1() {
|
|
let mut builder = SemanticTokensBuilder::new();
|
|
builder.push(2, 0, 5, 1, 1);
|
|
builder.push(2, 10, 1, 2, 2);
|
|
builder.push(2, 15, 2, 3, 3);
|
|
builder.push(1, 0, 4, 4, 4);
|
|
assert_eq!(
|
|
builder.build(None).data,
|
|
vec![
|
|
SemanticToken {
|
|
delta_line: 1,
|
|
delta_start: 0,
|
|
length: 4,
|
|
token_type: 4,
|
|
token_modifiers_bitset: 4
|
|
},
|
|
SemanticToken {
|
|
delta_line: 1,
|
|
delta_start: 0,
|
|
length: 5,
|
|
token_type: 1,
|
|
token_modifiers_bitset: 1
|
|
},
|
|
SemanticToken {
|
|
delta_line: 0,
|
|
delta_start: 10,
|
|
length: 1,
|
|
token_type: 2,
|
|
token_modifiers_bitset: 2
|
|
},
|
|
SemanticToken {
|
|
delta_line: 0,
|
|
delta_start: 5,
|
|
length: 2,
|
|
token_type: 3,
|
|
token_modifiers_bitset: 3
|
|
}
|
|
]
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn test_semantic_tokens_builder_out_of_order_2() {
|
|
let mut builder = SemanticTokensBuilder::new();
|
|
builder.push(2, 10, 5, 1, 1);
|
|
builder.push(2, 2, 4, 2, 2);
|
|
assert_eq!(
|
|
builder.build(None).data,
|
|
vec![
|
|
SemanticToken {
|
|
delta_line: 2,
|
|
delta_start: 2,
|
|
length: 4,
|
|
token_type: 2,
|
|
token_modifiers_bitset: 2
|
|
},
|
|
SemanticToken {
|
|
delta_line: 0,
|
|
delta_start: 8,
|
|
length: 5,
|
|
token_type: 1,
|
|
token_modifiers_bitset: 1
|
|
}
|
|
]
|
|
);
|
|
}
|
|
}
|