| 1 |
import {Input} from "@lezer/common" |
| 2 |
import {Stack} from "./stack" |
| 3 |
import {Seq} from "./constants" |
| 4 |
import {decodeArray} from "./decode" |
| 5 |
|
| 6 |
export class CachedToken { |
| 7 |
start = -1 |
| 8 |
value = -1 |
| 9 |
end = -1 |
| 10 |
extended = -1 |
| 11 |
lookAhead = 0 |
| 12 |
mask = 0 |
| 13 |
context = 0 |
| 14 |
} |
| 15 |
|
| 16 |
const nullToken = new CachedToken |
| 17 |
|
| 18 |
/// [Tokenizers](#lr.ExternalTokenizer) interact with the input |
| 19 |
/// through this interface. It presents the input as a stream of |
| 20 |
/// characters, tracking lookahead and hiding the complexity of |
| 21 |
/// [ranges](#common.Parser.parse^ranges) from tokenizer code. |
| 22 |
export class InputStream { |
| 23 |
/// @internal |
| 24 |
chunk = "" |
| 25 |
/// @internal |
| 26 |
chunkOff = 0 |
| 27 |
/// @internal |
| 28 |
chunkPos: number |
| 29 |
/// Backup chunk |
| 30 |
private chunk2 = "" |
| 31 |
private chunk2Pos = 0 |
| 32 |
|
| 33 |
/// The character code of the next code unit in the input, or -1 |
| 34 |
/// when the stream is at the end of the input. |
| 35 |
next: number = -1 |
| 36 |
|
| 37 |
/// @internal |
| 38 |
token = nullToken |
| 39 |
|
| 40 |
/// The current position of the stream. Note that, due to parses |
| 41 |
/// being able to cover non-contiguous |
| 42 |
/// [ranges](#common.Parser.startParse), advancing the stream does |
| 43 |
/// not always mean its position moves a single unit. |
| 44 |
pos: number |
| 45 |
|
| 46 |
/// @internal |
| 47 |
end: number |
| 48 |
|
| 49 |
private rangeIndex = 0 |
| 50 |
private range: {from: number, to: number} |
| 51 |
|
| 52 |
/// @internal |
| 53 |
constructor( |
| 54 |
/// @internal |
| 55 |
readonly input: Input, |
| 56 |
/// @internal |
| 57 |
readonly ranges: readonly {from: number, to: number}[] |
| 58 |
) { |
| 59 |
this.pos = this.chunkPos = ranges[0].from |
| 60 |
this.range = ranges[0] |
| 61 |
this.end = ranges[ranges.length - 1].to |
| 62 |
this.readNext() |
| 63 |
} |
| 64 |
|
| 65 |
/// @internal |
| 66 |
resolveOffset(offset: number, assoc: -1 | 1) { |
| 67 |
let range = this.range, index = this.rangeIndex |
| 68 |
let pos = this.pos + offset |
| 69 |
while (pos < range.from) { |
| 70 |
if (!index) return null |
| 71 |
let next = this.ranges[--index] |
| 72 |
pos -= range.from - next.to |
| 73 |
range = next |
| 74 |
} |
| 75 |
while (assoc < 0 ? pos > range.to : pos >= range.to) { |
| 76 |
if (index == this.ranges.length - 1) return null |
| 77 |
let next = this.ranges[++index] |
| 78 |
pos += next.from - range.to |
| 79 |
range = next |
| 80 |
} |
| 81 |
return pos |
| 82 |
} |
| 83 |
|
| 84 |
/// @internal |
| 85 |
clipPos(pos: number) { |
| 86 |
if (pos >= this.range.from && pos < this.range.to) return pos |
| 87 |
for (let range of this.ranges) if (range.to > pos) return Math.max(pos, range.from) |
| 88 |
return this.end |
| 89 |
} |
| 90 |
|
| 91 |
/// Look at a code unit near the stream position. `.peek(0)` equals |
| 92 |
/// `.next`, `.peek(-1)` gives you the previous character, and so |
| 93 |
/// on. |
| 94 |
/// |
| 95 |
/// Note that looking around during tokenizing creates dependencies |
| 96 |
/// on potentially far-away content, which may reduce the |
| 97 |
/// effectiveness incremental parsing—when looking forward—or even |
| 98 |
/// cause invalid reparses when looking backward more than 25 code |
| 99 |
/// units, since the library does not track lookbehind. |
| 100 |
peek(offset: number) { |
| 101 |
let idx = this.chunkOff + offset, pos, result |
| 102 |
if (idx >= 0 && idx < this.chunk.length) { |
| 103 |
pos = this.pos + offset |
| 104 |
result = this.chunk.charCodeAt(idx) |
| 105 |
} else { |
| 106 |
let resolved = this.resolveOffset(offset, 1) |
| 107 |
if (resolved == null) return -1 |
| 108 |
pos = resolved |
| 109 |
if (pos >= this.chunk2Pos && pos < this.chunk2Pos + this.chunk2.length) { |
| 110 |
result = this.chunk2.charCodeAt(pos - this.chunk2Pos) |
| 111 |
} else { |
| 112 |
let i = this.rangeIndex, range = this.range |
| 113 |
while (range.to <= pos) range = this.ranges[++i] |
| 114 |
this.chunk2 = this.input.chunk(this.chunk2Pos = pos) |
| 115 |
if (pos + this.chunk2.length > range.to) this.chunk2 = this.chunk2.slice(0, range.to - pos) |
| 116 |
result = this.chunk2.charCodeAt(0) |
| 117 |
} |
| 118 |
} |
| 119 |
if (pos >= this.token.lookAhead) this.token.lookAhead = pos + 1 |
| 120 |
return result |
| 121 |
} |
| 122 |
|
| 123 |
/// Accept a token. By default, the end of the token is set to the |
| 124 |
/// current stream position, but you can pass an offset (relative to |
| 125 |
/// the stream position) to change that. |
| 126 |
acceptToken(token: number, endOffset = 0) { |
| 127 |
let end = endOffset ? this.resolveOffset(endOffset, -1) : this.pos |
| 128 |
if (end == null || end < this.token.start) throw new RangeError("Token end out of bounds") |
| 129 |
this.token.value = token |
| 130 |
this.token.end = end |
| 131 |
} |
| 132 |
|
| 133 |
/// Accept a token ending at a specific given position. |
| 134 |
acceptTokenTo(token: number, endPos: number) { |
| 135 |
this.token.value = token |
| 136 |
this.token.end = endPos |
| 137 |
} |
| 138 |
|
| 139 |
private getChunk() { |
| 140 |
if (this.pos >= this.chunk2Pos && this.pos < this.chunk2Pos + this.chunk2.length) { |
| 141 |
let {chunk, chunkPos} = this |
| 142 |
this.chunk = this.chunk2; this.chunkPos = this.chunk2Pos |
| 143 |
this.chunk2 = chunk; this.chunk2Pos = chunkPos |
| 144 |
this.chunkOff = this.pos - this.chunkPos |
| 145 |
} else { |
| 146 |
this.chunk2 = this.chunk; this.chunk2Pos = this.chunkPos |
| 147 |
let nextChunk = this.input.chunk(this.pos) |
| 148 |
let end = this.pos + nextChunk.length |
| 149 |
this.chunk = end > this.range.to ? nextChunk.slice(0, this.range.to - this.pos) : nextChunk |
| 150 |
this.chunkPos = this.pos |
| 151 |
this.chunkOff = 0 |
| 152 |
} |
| 153 |
} |
| 154 |
|
| 155 |
private readNext() { |
| 156 |
if (this.chunkOff >= this.chunk.length) { |
| 157 |
this.getChunk() |
| 158 |
if (this.chunkOff == this.chunk.length) return this.next = -1 |
| 159 |
} |
| 160 |
return this.next = this.chunk.charCodeAt(this.chunkOff) |
| 161 |
} |
| 162 |
|
| 163 |
/// Move the stream forward N (defaults to 1) code units. Returns |
| 164 |
/// the new value of [`next`](#lr.InputStream.next). |
| 165 |
advance(n = 1) { |
| 166 |
this.chunkOff += n |
| 167 |
while (this.pos + n >= this.range.to) { |
| 168 |
if (this.rangeIndex == this.ranges.length - 1) return this.setDone() |
| 169 |
n -= this.range.to - this.pos |
| 170 |
this.range = this.ranges[++this.rangeIndex] |
| 171 |
this.pos = this.range.from |
| 172 |
} |
| 173 |
this.pos += n |
| 174 |
if (this.pos >= this.token.lookAhead) this.token.lookAhead = this.pos + 1 |
| 175 |
return this.readNext() |
| 176 |
} |
| 177 |
|
| 178 |
private setDone() { |
| 179 |
this.pos = this.chunkPos = this.end |
| 180 |
this.range = this.ranges[this.rangeIndex = this.ranges.length - 1] |
| 181 |
this.chunk = "" |
| 182 |
return this.next = -1 |
| 183 |
} |
| 184 |
|
| 185 |
/// @internal |
| 186 |
reset(pos: number, token?: CachedToken) { |
| 187 |
if (token) { |
| 188 |
this.token = token |
| 189 |
token.start = pos |
| 190 |
token.lookAhead = pos + 1 |
| 191 |
token.value = token.extended = -1 |
| 192 |
} else { |
| 193 |
this.token = nullToken |
| 194 |
} |
| 195 |
if (this.pos != pos) { |
| 196 |
this.pos = pos |
| 197 |
if (pos == this.end) { |
| 198 |
this.setDone() |
| 199 |
return this |
| 200 |
} |
| 201 |
while (pos < this.range.from) this.range = this.ranges[--this.rangeIndex] |
| 202 |
while (pos >= this.range.to) this.range = this.ranges[++this.rangeIndex] |
| 203 |
if (pos >= this.chunkPos && pos < this.chunkPos + this.chunk.length) { |
| 204 |
this.chunkOff = pos - this.chunkPos |
| 205 |
} else { |
| 206 |
this.chunk = "" |
| 207 |
this.chunkOff = 0 |
| 208 |
} |
| 209 |
this.readNext() |
| 210 |
} |
| 211 |
return this |
| 212 |
} |
| 213 |
|
| 214 |
/// @internal |
| 215 |
read(from: number, to: number) { |
| 216 |
if (from >= this.chunkPos && to <= this.chunkPos + this.chunk.length) |
| 217 |
return this.chunk.slice(from - this.chunkPos, to - this.chunkPos) |
| 218 |
if (from >= this.chunk2Pos && to <= this.chunk2Pos + this.chunk2.length) |
| 219 |
return this.chunk2.slice(from - this.chunk2Pos, to - this.chunk2Pos) |
| 220 |
if (from >= this.range.from && to <= this.range.to) |
| 221 |
return this.input.read(from, to) |
| 222 |
let result = "" |
| 223 |
for (let r of this.ranges) { |
| 224 |
if (r.from >= to) break |
| 225 |
if (r.to > from) result += this.input.read(Math.max(r.from, from), Math.min(r.to, to)) |
| 226 |
} |
| 227 |
return result |
| 228 |
} |
| 229 |
} |
| 230 |
|
| 231 |
export interface Tokenizer { |
| 232 |
/// @internal |
| 233 |
token(input: InputStream, stack: Stack): void |
| 234 |
/// @internal |
| 235 |
contextual: boolean |
| 236 |
/// @internal |
| 237 |
fallback: boolean |
| 238 |
/// @internal |
| 239 |
extend: boolean |
| 240 |
} |
| 241 |
|
| 242 |
/// @internal |
| 243 |
export class TokenGroup implements Tokenizer { |
| 244 |
contextual!: boolean |
| 245 |
fallback!: boolean |
| 246 |
extend!: boolean |
| 247 |
|
| 248 |
constructor(readonly data: Readonly<Uint16Array>, readonly id: number) {} |
| 249 |
|
| 250 |
token(input: InputStream, stack: Stack) { |
| 251 |
let {parser} = stack.p |
| 252 |
readToken(this.data, input, stack, this.id, parser.data, parser.tokenPrecTable) |
| 253 |
} |
| 254 |
} |
| 255 |
|
| 256 |
TokenGroup.prototype.contextual = TokenGroup.prototype.fallback = TokenGroup.prototype.extend = false |
| 257 |
|
| 258 |
/// @hide |
| 259 |
export class LocalTokenGroup implements Tokenizer { |
| 260 |
contextual!: boolean |
| 261 |
fallback!: boolean |
| 262 |
extend!: boolean |
| 263 |
readonly data: Readonly<Uint16Array> |
| 264 |
|
| 265 |
constructor(data: Readonly<Uint16Array> | string, readonly precTable: number, readonly elseToken?: number) { |
| 266 |
this.data = typeof data == "string" ? decodeArray<Readonly<Uint16Array>>(data) : data |
| 267 |
} |
| 268 |
|
| 269 |
token(input: InputStream, stack: Stack) { |
| 270 |
let start = input.pos, skipped = 0 |
| 271 |
for (;;) { |
| 272 |
let atEof = input.next < 0, nextPos = input.resolveOffset(1, 1) |
| 273 |
readToken(this.data, input, stack, 0, this.data, this.precTable) |
| 274 |
if (input.token.value > -1) break |
| 275 |
if (this.elseToken == null) return |
| 276 |
if (!atEof) skipped++ |
| 277 |
if (nextPos == null) break |
| 278 |
input.reset(nextPos, input.token) |
| 279 |
} |
| 280 |
if (skipped) { |
| 281 |
input.reset(start, input.token) |
| 282 |
input.acceptToken(this.elseToken!, skipped) |
| 283 |
} |
| 284 |
} |
| 285 |
} |
| 286 |
|
| 287 |
LocalTokenGroup.prototype.contextual = TokenGroup.prototype.fallback = TokenGroup.prototype.extend = false |
| 288 |
|
| 289 |
interface ExternalOptions { |
| 290 |
/// When set to true, mark this tokenizer as depending on the |
| 291 |
/// current parse stack, which prevents its result from being cached |
| 292 |
/// between parser actions at the same positions. |
| 293 |
contextual?: boolean, |
| 294 |
/// By defaults, when a tokenizer returns a token, that prevents |
| 295 |
/// tokenizers with lower precedence from even running. When |
| 296 |
/// `fallback` is true, the tokenizer is allowed to run when a |
| 297 |
/// previous tokenizer returned a token that didn't match any of the |
| 298 |
/// current state's actions. |
| 299 |
fallback?: boolean |
| 300 |
/// When set to true, tokenizing will not stop after this tokenizer |
| 301 |
/// has produced a token. (But it will still fail to reach this one |
| 302 |
/// if a higher-precedence tokenizer produced a token.) |
| 303 |
extend?: boolean |
| 304 |
} |
| 305 |
|
| 306 |
/// `@external tokens` declarations in the grammar should resolve to |
| 307 |
/// an instance of this class. |
| 308 |
export class ExternalTokenizer { |
| 309 |
/// @internal |
| 310 |
contextual: boolean |
| 311 |
/// @internal |
| 312 |
fallback: boolean |
| 313 |
/// @internal |
| 314 |
extend: boolean |
| 315 |
|
| 316 |
/// Create a tokenizer. The first argument is the function that, |
| 317 |
/// given an input stream, scans for the types of tokens it |
| 318 |
/// recognizes at the stream's position, and calls |
| 319 |
/// [`acceptToken`](#lr.InputStream.acceptToken) when it finds |
| 320 |
/// one. |
| 321 |
constructor( |
| 322 |
/// @internal |
| 323 |
readonly token: (input: InputStream, stack: Stack) => void, |
| 324 |
options: ExternalOptions = {} |
| 325 |
) { |
| 326 |
this.contextual = !!options.contextual |
| 327 |
this.fallback = !!options.fallback |
| 328 |
this.extend = !!options.extend |
| 329 |
} |
| 330 |
} |
| 331 |
|
| 332 |
// Tokenizer data is stored a big uint16 array containing, for each |
| 333 |
// state: |
| 334 |
// |
| 335 |
// - A group bitmask, indicating what token groups are reachable from |
| 336 |
// this state, so that paths that can only lead to tokens not in |
| 337 |
// any of the current groups can be cut off early. |
| 338 |
// |
| 339 |
// - The position of the end of the state's sequence of accepting |
| 340 |
// tokens |
| 341 |
// |
| 342 |
// - The number of outgoing edges for the state |
| 343 |
// |
| 344 |
// - The accepting tokens, as (token id, group mask) pairs |
| 345 |
// |
| 346 |
// - The outgoing edges, as (start character, end character, state |
| 347 |
// index) triples, with end character being exclusive |
| 348 |
// |
| 349 |
// This function interprets that data, running through a stream as |
| 350 |
// long as new states with the a matching group mask can be reached, |
| 351 |
// and updating `input.token` when it matches a token. |
| 352 |
function readToken(data: Readonly<Uint16Array>, |
| 353 |
input: InputStream, |
| 354 |
stack: Stack, |
| 355 |
group: number, |
| 356 |
precTable: Readonly<Uint16Array>, |
| 357 |
precOffset: number) { |
| 358 |
let state = 0, groupMask = 1 << group, {dialect} = stack.p.parser |
| 359 |
scan: for (;;) { |
| 360 |
if ((groupMask & data[state]) == 0) break |
| 361 |
let accEnd = data[state + 1] |
| 362 |
// Check whether this state can lead to a token in the current group |
| 363 |
// Accept tokens in this state, possibly overwriting |
| 364 |
// lower-precedence / shorter tokens |
| 365 |
for (let i = state + 3; i < accEnd; i += 2) if ((data[i + 1] & groupMask) > 0) { |
| 366 |
let term = data[i] |
| 367 |
if (dialect.allows(term) && |
| 368 |
(input.token.value == -1 || input.token.value == term || |
| 369 |
overrides(term, input.token.value, precTable, precOffset))) { |
| 370 |
input.acceptToken(term) |
| 371 |
break |
| 372 |
} |
| 373 |
} |
| 374 |
let next = input.next, low = 0, high = data[state + 2] |
| 375 |
// Special case for EOF |
| 376 |
if (input.next < 0 && high > low && data[accEnd + high * 3 - 3] == Seq.End) { |
| 377 |
state = data[accEnd + high * 3 - 1] |
| 378 |
continue scan |
| 379 |
} |
| 380 |
// Do a binary search on the state's edges |
| 381 |
for (; low < high;) { |
| 382 |
let mid = (low + high) >> 1 |
| 383 |
let index = accEnd + mid + (mid << 1) |
| 384 |
let from = data[index], to = data[index + 1] || 0x10000 |
| 385 |
if (next < from) high = mid |
| 386 |
else if (next >= to) low = mid + 1 |
| 387 |
else { state = data[index + 2]; input.advance(); continue scan } |
| 388 |
} |
| 389 |
break |
| 390 |
} |
| 391 |
} |
| 392 |
|
| 393 |
function findOffset(data: Readonly<Uint16Array>, start: number, term: number) { |
| 394 |
for (let i = start, next; (next = data[i]) != Seq.End; i++) |
| 395 |
if (next == term) return i - start |
| 396 |
return -1 |
| 397 |
} |
| 398 |
|
| 399 |
function overrides(token: number, prev: number, tableData: Readonly<Uint16Array>, tableOffset: number) { |
| 400 |
let iPrev = findOffset(tableData, tableOffset, prev) |
| 401 |
return iPrev < 0 || findOffset(tableData, tableOffset, token) < iPrev |
| 402 |
} |