the-forest/client/node_modules/sucrase/dist/esm/TokenProcessor.js
2024-09-17 20:35:18 -04:00

358 lines
9.9 KiB
JavaScript

import { TokenType as tt} from "./parser/tokenizer/types";
import isAsyncOperation from "./util/isAsyncOperation";
export default class TokenProcessor {
__init() {this.resultCode = ""}
// Array mapping input token index to optional string index position in the
// output code.
__init2() {this.resultMappings = new Array(this.tokens.length)}
__init3() {this.tokenIndex = 0}
constructor(
code,
tokens,
isFlowEnabled,
disableESTransforms,
helperManager,
) {;this.code = code;this.tokens = tokens;this.isFlowEnabled = isFlowEnabled;this.disableESTransforms = disableESTransforms;this.helperManager = helperManager;TokenProcessor.prototype.__init.call(this);TokenProcessor.prototype.__init2.call(this);TokenProcessor.prototype.__init3.call(this);}
/**
* Snapshot the token state in a way that can be restored later, useful for
* things like lookahead.
*
* resultMappings do not need to be copied since in all use cases, they will
* be overwritten anyway after restore.
*/
snapshot() {
return {
resultCode: this.resultCode,
tokenIndex: this.tokenIndex,
};
}
restoreToSnapshot(snapshot) {
this.resultCode = snapshot.resultCode;
this.tokenIndex = snapshot.tokenIndex;
}
/**
* Remove and return the code generated since the snapshot, leaving the
* current token position in-place. Unlike most TokenProcessor operations,
* this operation can result in input/output line number mismatches because
* the removed code may contain newlines, so this operation should be used
* sparingly.
*/
dangerouslyGetAndRemoveCodeSinceSnapshot(snapshot) {
const result = this.resultCode.slice(snapshot.resultCode.length);
this.resultCode = snapshot.resultCode;
return result;
}
reset() {
this.resultCode = "";
this.resultMappings = new Array(this.tokens.length);
this.tokenIndex = 0;
}
matchesContextualAtIndex(index, contextualKeyword) {
return (
this.matches1AtIndex(index, tt.name) &&
this.tokens[index].contextualKeyword === contextualKeyword
);
}
identifierNameAtIndex(index) {
// TODO: We need to process escapes since technically you can have unicode escapes in variable
// names.
return this.identifierNameForToken(this.tokens[index]);
}
identifierNameAtRelativeIndex(relativeIndex) {
return this.identifierNameForToken(this.tokenAtRelativeIndex(relativeIndex));
}
identifierName() {
return this.identifierNameForToken(this.currentToken());
}
identifierNameForToken(token) {
return this.code.slice(token.start, token.end);
}
rawCodeForToken(token) {
return this.code.slice(token.start, token.end);
}
stringValueAtIndex(index) {
return this.stringValueForToken(this.tokens[index]);
}
stringValue() {
return this.stringValueForToken(this.currentToken());
}
stringValueForToken(token) {
// This is used to identify when two imports are the same and to resolve TypeScript enum keys.
// Ideally we'd process escapes within the strings, but for now we pretty much take the raw
// code.
return this.code.slice(token.start + 1, token.end - 1);
}
matches1AtIndex(index, t1) {
return this.tokens[index].type === t1;
}
matches2AtIndex(index, t1, t2) {
return this.tokens[index].type === t1 && this.tokens[index + 1].type === t2;
}
matches3AtIndex(index, t1, t2, t3) {
return (
this.tokens[index].type === t1 &&
this.tokens[index + 1].type === t2 &&
this.tokens[index + 2].type === t3
);
}
matches1(t1) {
return this.tokens[this.tokenIndex].type === t1;
}
matches2(t1, t2) {
return this.tokens[this.tokenIndex].type === t1 && this.tokens[this.tokenIndex + 1].type === t2;
}
matches3(t1, t2, t3) {
return (
this.tokens[this.tokenIndex].type === t1 &&
this.tokens[this.tokenIndex + 1].type === t2 &&
this.tokens[this.tokenIndex + 2].type === t3
);
}
matches4(t1, t2, t3, t4) {
return (
this.tokens[this.tokenIndex].type === t1 &&
this.tokens[this.tokenIndex + 1].type === t2 &&
this.tokens[this.tokenIndex + 2].type === t3 &&
this.tokens[this.tokenIndex + 3].type === t4
);
}
matches5(t1, t2, t3, t4, t5) {
return (
this.tokens[this.tokenIndex].type === t1 &&
this.tokens[this.tokenIndex + 1].type === t2 &&
this.tokens[this.tokenIndex + 2].type === t3 &&
this.tokens[this.tokenIndex + 3].type === t4 &&
this.tokens[this.tokenIndex + 4].type === t5
);
}
matchesContextual(contextualKeyword) {
return this.matchesContextualAtIndex(this.tokenIndex, contextualKeyword);
}
matchesContextIdAndLabel(type, contextId) {
return this.matches1(type) && this.currentToken().contextId === contextId;
}
previousWhitespaceAndComments() {
let whitespaceAndComments = this.code.slice(
this.tokenIndex > 0 ? this.tokens[this.tokenIndex - 1].end : 0,
this.tokenIndex < this.tokens.length ? this.tokens[this.tokenIndex].start : this.code.length,
);
if (this.isFlowEnabled) {
whitespaceAndComments = whitespaceAndComments.replace(/@flow/g, "");
}
return whitespaceAndComments;
}
replaceToken(newCode) {
this.resultCode += this.previousWhitespaceAndComments();
this.appendTokenPrefix();
this.resultMappings[this.tokenIndex] = this.resultCode.length;
this.resultCode += newCode;
this.appendTokenSuffix();
this.tokenIndex++;
}
replaceTokenTrimmingLeftWhitespace(newCode) {
this.resultCode += this.previousWhitespaceAndComments().replace(/[^\r\n]/g, "");
this.appendTokenPrefix();
this.resultMappings[this.tokenIndex] = this.resultCode.length;
this.resultCode += newCode;
this.appendTokenSuffix();
this.tokenIndex++;
}
removeInitialToken() {
this.replaceToken("");
}
removeToken() {
this.replaceTokenTrimmingLeftWhitespace("");
}
/**
* Remove all code until the next }, accounting for balanced braces.
*/
removeBalancedCode() {
let braceDepth = 0;
while (!this.isAtEnd()) {
if (this.matches1(tt.braceL)) {
braceDepth++;
} else if (this.matches1(tt.braceR)) {
if (braceDepth === 0) {
return;
}
braceDepth--;
}
this.removeToken();
}
}
copyExpectedToken(tokenType) {
if (this.tokens[this.tokenIndex].type !== tokenType) {
throw new Error(`Expected token ${tokenType}`);
}
this.copyToken();
}
copyToken() {
this.resultCode += this.previousWhitespaceAndComments();
this.appendTokenPrefix();
this.resultMappings[this.tokenIndex] = this.resultCode.length;
this.resultCode += this.code.slice(
this.tokens[this.tokenIndex].start,
this.tokens[this.tokenIndex].end,
);
this.appendTokenSuffix();
this.tokenIndex++;
}
copyTokenWithPrefix(prefix) {
this.resultCode += this.previousWhitespaceAndComments();
this.appendTokenPrefix();
this.resultCode += prefix;
this.resultMappings[this.tokenIndex] = this.resultCode.length;
this.resultCode += this.code.slice(
this.tokens[this.tokenIndex].start,
this.tokens[this.tokenIndex].end,
);
this.appendTokenSuffix();
this.tokenIndex++;
}
appendTokenPrefix() {
const token = this.currentToken();
if (token.numNullishCoalesceStarts || token.isOptionalChainStart) {
token.isAsyncOperation = isAsyncOperation(this);
}
if (this.disableESTransforms) {
return;
}
if (token.numNullishCoalesceStarts) {
for (let i = 0; i < token.numNullishCoalesceStarts; i++) {
if (token.isAsyncOperation) {
this.resultCode += "await ";
this.resultCode += this.helperManager.getHelperName("asyncNullishCoalesce");
} else {
this.resultCode += this.helperManager.getHelperName("nullishCoalesce");
}
this.resultCode += "(";
}
}
if (token.isOptionalChainStart) {
if (token.isAsyncOperation) {
this.resultCode += "await ";
}
if (this.tokenIndex > 0 && this.tokenAtRelativeIndex(-1).type === tt._delete) {
if (token.isAsyncOperation) {
this.resultCode += this.helperManager.getHelperName("asyncOptionalChainDelete");
} else {
this.resultCode += this.helperManager.getHelperName("optionalChainDelete");
}
} else if (token.isAsyncOperation) {
this.resultCode += this.helperManager.getHelperName("asyncOptionalChain");
} else {
this.resultCode += this.helperManager.getHelperName("optionalChain");
}
this.resultCode += "([";
}
}
appendTokenSuffix() {
const token = this.currentToken();
if (token.isOptionalChainEnd && !this.disableESTransforms) {
this.resultCode += "])";
}
if (token.numNullishCoalesceEnds && !this.disableESTransforms) {
for (let i = 0; i < token.numNullishCoalesceEnds; i++) {
this.resultCode += "))";
}
}
}
appendCode(code) {
this.resultCode += code;
}
currentToken() {
return this.tokens[this.tokenIndex];
}
currentTokenCode() {
const token = this.currentToken();
return this.code.slice(token.start, token.end);
}
tokenAtRelativeIndex(relativeIndex) {
return this.tokens[this.tokenIndex + relativeIndex];
}
currentIndex() {
return this.tokenIndex;
}
/**
* Move to the next token. Only suitable in preprocessing steps. When
* generating new code, you should use copyToken or removeToken.
*/
nextToken() {
if (this.tokenIndex === this.tokens.length) {
throw new Error("Unexpectedly reached end of input.");
}
this.tokenIndex++;
}
previousToken() {
this.tokenIndex--;
}
finish() {
if (this.tokenIndex !== this.tokens.length) {
throw new Error("Tried to finish processing tokens before reaching the end.");
}
this.resultCode += this.previousWhitespaceAndComments();
return {code: this.resultCode, mappings: this.resultMappings};
}
isAtEnd() {
return this.tokenIndex === this.tokens.length;
}
}