Skip to content

Commit 23a5714

Browse files
legendecaslegendecas
committed
lib,src: refactor assert to load error source from memory
The source code is available from V8 API and assert can avoid reading the source file from the filesystem and parse the file again.
1 parent 5b32bb1 commit 23a5714

File tree

5 files changed

+181
-257
lines changed

5 files changed

+181
-257
lines changed

‎lib/internal/assert/utils.js

Lines changed: 9 additions & 205 deletions
Original file line numberDiff line numberDiff line change
@@ -1,38 +1,21 @@
11
'use strict';
22

33
const {
4-
ArrayPrototypeShift,
54
Error,
65
ErrorCaptureStackTrace,
7-
FunctionPrototypeBind,
8-
RegExpPrototypeSymbolReplace,
9-
SafeMap,
106
StringPrototypeCharCodeAt,
11-
StringPrototypeIncludes,
127
StringPrototypeReplace,
13-
StringPrototypeSlice,
14-
StringPrototypeSplit,
15-
StringPrototypeStartsWith,
168
} = primordials;
179

18-
const { Buffer } = require('buffer');
1910
const {
2011
isErrorStackTraceLimitWritable,
21-
overrideStackTrace,
2212
} = require('internal/errors');
2313
const AssertionError = require('internal/assert/assertion_error');
24-
const { openSync, closeSync, readSync } = require('fs');
25-
const { EOL } = require('internal/constants');
26-
const { BuiltinModule } = require('internal/bootstrap/realm');
2714
const { isError } = require('internal/util');
2815

29-
const errorCache = new SafeMap();
30-
const { fileURLToPath } = require('internal/url');
31-
32-
let parseExpressionAt;
33-
let findNodeAround;
34-
let tokenizer;
35-
let decoder;
16+
const {
17+
getErrorSourceExpression,
18+
} = require('internal/errors/error_source');
3619

3720
// Escape control characters but not \n and \t to keep the line breaks and
3821
// indentation intact.
@@ -50,111 +33,7 @@ const meta = [
5033

5134
const escapeFn = (str) => meta[StringPrototypeCharCodeAt(str, 0)];
5235

53-
function findColumn(fd, column, code) {
54-
if (code.length > column + 100) {
55-
try {
56-
return parseCode(code, column);
57-
} catch {
58-
// End recursion in case no code could be parsed. The expression should
59-
// have been found after 2500 characters, so stop trying.
60-
if (code.length - column > 2500) {
61-
// eslint-disable-next-line no-throw-literal
62-
throw null;
63-
}
64-
}
65-
}
66-
// Read up to 2500 bytes more than necessary in columns. That way we address
67-
// multi byte characters and read enough data to parse the code.
68-
const bytesToRead = column - code.length + 2500;
69-
const buffer = Buffer.allocUnsafe(bytesToRead);
70-
const bytesRead = readSync(fd, buffer, 0, bytesToRead);
71-
code += decoder.write(buffer.slice(0, bytesRead));
72-
// EOF: fast path.
73-
if (bytesRead < bytesToRead) {
74-
return parseCode(code, column);
75-
}
76-
// Read potentially missing code.
77-
return findColumn(fd, column, code);
78-
}
79-
80-
function getCode(fd, line, column) {
81-
let bytesRead = 0;
82-
if (line === 0) {
83-
// Special handle line number one. This is more efficient and simplifies the
84-
// rest of the algorithm. Read more than the regular column number in bytes
85-
// to prevent multiple reads in case multi byte characters are used.
86-
return findColumn(fd, column, '');
87-
}
88-
let lines = 0;
89-
// Prevent blocking the event loop by limiting the maximum amount of
90-
// data that may be read.
91-
let maxReads = 32; // bytesPerRead * maxReads = 512 KiB
92-
const bytesPerRead = 16384;
93-
// Use a single buffer up front that is reused until the call site is found.
94-
let buffer = Buffer.allocUnsafe(bytesPerRead);
95-
while (maxReads-- !== 0) {
96-
// Only allocate a new buffer in case the needed line is found. All data
97-
// before that can be discarded.
98-
buffer = lines < line ? buffer : Buffer.allocUnsafe(bytesPerRead);
99-
bytesRead = readSync(fd, buffer, 0, bytesPerRead);
100-
// Read the buffer until the required code line is found.
101-
for (let i = 0; i < bytesRead; i++) {
102-
if (buffer[i] === 10 && ++lines === line) {
103-
// If the end of file is reached, directly parse the code and return.
104-
if (bytesRead < bytesPerRead) {
105-
return parseCode(buffer.toString('utf8', i + 1, bytesRead), column);
106-
}
107-
// Check if the read code is sufficient or read more until the whole
108-
// expression is read. Make sure multi byte characters are preserved
109-
// properly by using the decoder.
110-
const code = decoder.write(buffer.slice(i + 1, bytesRead));
111-
return findColumn(fd, column, code);
112-
}
113-
}
114-
}
115-
}
116-
117-
function parseCode(code, offset) {
118-
// Lazy load acorn.
119-
if (parseExpressionAt === undefined) {
120-
const Parser = require('internal/deps/acorn/acorn/dist/acorn').Parser;
121-
({ findNodeAround } = require('internal/deps/acorn/acorn-walk/dist/walk'));
122-
123-
parseExpressionAt = FunctionPrototypeBind(Parser.parseExpressionAt, Parser);
124-
tokenizer = FunctionPrototypeBind(Parser.tokenizer, Parser);
125-
}
126-
let node;
127-
let start;
128-
// Parse the read code until the correct expression is found.
129-
for (const token of tokenizer(code, { ecmaVersion: 'latest' })) {
130-
start = token.start;
131-
if (start > offset) {
132-
// No matching expression found. This could happen if the assert
133-
// expression is bigger than the provided buffer.
134-
break;
135-
}
136-
try {
137-
node = parseExpressionAt(code, start, { ecmaVersion: 'latest' });
138-
// Find the CallExpression in the tree.
139-
node = findNodeAround(node, offset, 'CallExpression');
140-
if (node?.node.end >= offset) {
141-
return [
142-
node.node.start,
143-
StringPrototypeReplace(StringPrototypeSlice(code,
144-
node.node.start, node.node.end),
145-
escapeSequencesRegExp, escapeFn),
146-
];
147-
}
148-
// eslint-disable-next-line no-unused-vars
149-
} catch (err) {
150-
continue;
151-
}
152-
}
153-
// eslint-disable-next-line no-throw-literal
154-
throw null;
155-
}
156-
157-
function getErrMessage(message, fn) {
36+
function getErrMessage(fn) {
15837
const tmpLimit = Error.stackTraceLimit;
15938
const errorStackTraceLimitIsWritable = isErrorStackTraceLimitWritable();
16039
// Make sure the limit is set to 1. Otherwise it could fail (<= 0) or it
@@ -166,85 +45,10 @@ function getErrMessage(message, fn) {
16645
ErrorCaptureStackTrace(err, fn);
16746
if (errorStackTraceLimitIsWritable) Error.stackTraceLimit = tmpLimit;
16847

169-
overrideStackTrace.set(err, (_, stack) => stack);
170-
const call = err.stack[0];
171-
172-
let filename = call.getFileName();
173-
const line = call.getLineNumber() - 1;
174-
let column = call.getColumnNumber() - 1;
175-
let identifier;
176-
177-
if (filename) {
178-
identifier = `${filename}${line}${column}`;
179-
180-
// Skip Node.js modules!
181-
if (StringPrototypeStartsWith(filename, 'node:') &&
182-
BuiltinModule.exists(StringPrototypeSlice(filename, 5))) {
183-
errorCache.set(identifier, undefined);
184-
return;
185-
}
186-
} else {
187-
return message;
188-
}
189-
190-
if (errorCache.has(identifier)) {
191-
return errorCache.get(identifier);
192-
}
193-
194-
let fd;
195-
try {
196-
// Set the stack trace limit to zero. This makes sure unexpected token
197-
// errors are handled faster.
198-
if (errorStackTraceLimitIsWritable) Error.stackTraceLimit = 0;
199-
200-
if (decoder === undefined) {
201-
const { StringDecoder } = require('string_decoder');
202-
decoder = new StringDecoder('utf8');
203-
}
204-
205-
// ESM file prop is a file proto. Convert that to path.
206-
// This ensure opensync will not throw ENOENT for ESM files.
207-
const fileProtoPrefix = 'file://';
208-
if (StringPrototypeStartsWith(filename, fileProtoPrefix)) {
209-
filename = fileURLToPath(filename);
210-
}
211-
212-
fd = openSync(filename, 'r', 0o666);
213-
// Reset column and message.
214-
({ 0: column, 1: message } = getCode(fd, line, column));
215-
// Flush unfinished multi byte characters.
216-
decoder.end();
217-
218-
// Always normalize indentation, otherwise the message could look weird.
219-
if (StringPrototypeIncludes(message, '\n')) {
220-
if (EOL === '\r\n') {
221-
message = RegExpPrototypeSymbolReplace(/\r\n/g, message, '\n');
222-
}
223-
const frames = StringPrototypeSplit(message, '\n');
224-
message = ArrayPrototypeShift(frames);
225-
for (let i = 0; i < frames.length; i++) {
226-
const frame = frames[i];
227-
let pos = 0;
228-
while (pos < column && (frame[pos] === ' ' || frame[pos] === '\t')) {
229-
pos++;
230-
}
231-
message += `\n ${StringPrototypeSlice(frame, pos)}`;
232-
}
233-
}
234-
message = `The expression evaluated to a falsy value:\n\n ${message}\n`;
235-
// Make sure to always set the cache! No matter if the message is
236-
// undefined or not
237-
errorCache.set(identifier, message);
238-
239-
return message;
240-
} catch {
241-
// Invalidate cache to prevent trying to read this part again.
242-
errorCache.set(identifier, undefined);
243-
} finally {
244-
// Reset limit.
245-
if (errorStackTraceLimitIsWritable) Error.stackTraceLimit = tmpLimit;
246-
if (fd !== undefined)
247-
closeSync(fd);
48+
let source = getErrorSourceExpression(err);
49+
if (source) {
50+
source = StringPrototypeReplace(source, escapeSequencesRegExp, escapeFn);
51+
return `The expression evaluated to a falsy value:\n\n ${source}\n`;
24852
}
24953
}
25054

@@ -257,7 +61,7 @@ function innerOk(fn, argLen, value, message) {
25761
message = 'No value argument passed to `assert.ok()`';
25862
} else if (message == null) {
25963
generatedMessage = true;
260-
message = getErrMessage(message, fn);
64+
message = getErrMessage(fn);
26165
} else if (isError(message)) {
26266
throw message;
26367
}

‎lib/internal/errors/error_source.js

Lines changed: 125 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,125 @@
1+
'use strict';
2+
3+
const {
4+
FunctionPrototypeBind,
5+
StringPrototypeSlice,
6+
} = primordials;
7+
8+
const {
9+
getErrorSourcePositions,
10+
} = internalBinding('errors');
11+
12+
/**
13+
* Get the source location of an error.
14+
*
15+
* The `error.stack` must not have been accessed. The resolution is based on the structured
16+
* error stack data.
17+
* @param {Error|object} error An error object, or an object being invoked with ErrorCaptureStackTrace
18+
* @returns {{sourceLine: string, startColumn: number}|undefined}
19+
*/
20+
function getErrorSourceLocation(error) {
21+
const pos = getErrorSourcePositions(error);
22+
const {
23+
sourceLine,
24+
startColumn,
25+
} = pos;
26+
27+
return { sourceLine, startColumn };
28+
}
29+
30+
const memberAccessTokens = [ '.', '?.', '[', ']' ];
31+
const memberNameTokens = [ 'name', 'string', 'num' ];
32+
let tokenizer;
33+
/**
34+
* Get the first expression in a code string at the startColumn.
35+
* @param {string} code source code line
36+
* @param {number} startColumn which column the error is constructed
37+
* @returns {string}
38+
*/
39+
function getFirstExpression(code, startColumn) {
40+
// Lazy load acorn.
41+
if (tokenizer === undefined) {
42+
const Parser = require('internal/deps/acorn/acorn/dist/acorn').Parser;
43+
tokenizer = FunctionPrototypeBind(Parser.tokenizer, Parser);
44+
}
45+
46+
let lastToken;
47+
let firstMemberAccessNameToken;
48+
let terminatingCol;
49+
let parenLvl = 0;
50+
// Tokenize the line to locate the expression at the startColumn.
51+
// The source line may be an incomplete JavaScript source, so do not parse the source line.
52+
for (const token of tokenizer(code, { ecmaVersion: 'latest' })) {
53+
// Peek before the startColumn.
54+
if (token.start < startColumn) {
55+
// There is a semicolon. This is a statement before the startColumn, so reset the memo.
56+
if (token.type.label === ';') {
57+
firstMemberAccessNameToken = null;
58+
continue;
59+
}
60+
// Try to memo the member access expressions before the startColumn, so that the
61+
// returned source code contains more info:
62+
// assert.ok(value)
63+
// ^
64+
// The member expression can also be like
65+
// assert['ok'](value) or assert?.ok(value)
66+
// ^ ^
67+
if (memberAccessTokens.includes(token.type.label) && lastToken?.type.label === 'name') {
68+
// First member access name token must be a 'name'.
69+
firstMemberAccessNameToken ??= lastToken;
70+
} else if (!memberAccessTokens.includes(token.type.label) &&
71+
!memberNameTokens.includes(token.type.label)) {
72+
// Reset the memo if it is not a simple member access.
73+
// For example: assert[(() => 'ok')](value)
74+
// ^
75+
firstMemberAccessNameToken = null;
76+
}
77+
lastToken = token;
78+
continue;
79+
}
80+
// Now after the startColumn, this must be an expression.
81+
if (token.type.label === '(') {
82+
parenLvl++;
83+
}
84+
if (token.type.label === ')') {
85+
parenLvl--;
86+
if (parenLvl === 0) {
87+
// A matched closing parenthesis found after the startColumn,
88+
// terminate here. Include the token.
89+
terminatingCol = token.start + 1;
90+
break;
91+
}
92+
}
93+
if (token.type.label === ';') {
94+
// A semicolon found after the startColumn, terminate here.
95+
terminatingCol = token;
96+
break;
97+
}
98+
// If no semicolon found after the startColumn. The string after the
99+
// startColumn must be the expression.
100+
}
101+
const start = firstMemberAccessNameToken?.start ?? startColumn;
102+
return StringPrototypeSlice(code, start, terminatingCol);
103+
}
104+
105+
/**
106+
* Get the source expression of an error.
107+
*
108+
* The `error.stack` must not have been accessed, or the source location may be incorrect. The
109+
* resolution is based on the structured error stack data.
110+
* @param {Error|object} error An error object, or an object being invoked with ErrorCaptureStackTrace
111+
* @returns {string|undefined}
112+
*/
113+
function getErrorSourceExpression(error) {
114+
const loc = getErrorSourceLocation(error);
115+
if (loc === undefined) {
116+
return;
117+
}
118+
const { sourceLine, startColumn } = loc;
119+
return getFirstExpression(sourceLine, startColumn);
120+
}
121+
122+
module.exports = {
123+
getErrorSourceLocation,
124+
getErrorSourceExpression,
125+
};

0 commit comments

Comments
 (0)