performance: Safely ignore error stack trace
I saw this library https://github.com/isaacs/catcher and I decided to wrote a benchmark:
Doing some changes on bench/error.js, the result was:
name | ops/sec | samples |
---|---|---|
Error | 337,720 | 64 |
Error (stackTraceLimit=0) | 3,284,296 | 94 |
NodeError | 283,026 | 99 |
NodeError (stackTraceLimit=0) | 3,280,933 | 95 |
NodeError Range | 214,825 | 92 |
NodeError Range (stackTraceLimit=0) | 3,313,990 | 99 |
Code
const { createBenchmarkSuite } = require('../common')
const suite = createBenchmarkSuite('Node.js Error')
suite
.add('Error', function () {
try {
new Error('test')
} catch (e) { }
})
.add('Error (stackTraceLimit=0)', function () {
const originalStackTraceLimit = Error.stackTraceLimit
Error.stackTraceLimit = 0
try {
new Error('test')
} catch (e) { }
finally {
Error.stackTraceLimit = originalStackTraceLimit;
}
})
.add('NodeError', function () {
try {
new TypeError('test')
} catch (e) { }
})
.add('NodeError (stackTraceLimit=0)', function () {
const originalStackTraceLimit = Error.stackTraceLimit
Error.stackTraceLimit = 0
try {
new TypeError('test')
} catch (e) { }
finally {
Error.stackTraceLimit = originalStackTraceLimit;
}
})
.add('NodeError Range', function () {
try {
new RangeError('test')
} catch (e) { }
})
.add('NodeError Range (stackTraceLimit=0)', function () {
const originalStackTraceLimit = Error.stackTraceLimit
Error.stackTraceLimit = 0
try {
new RangeError('test')
} catch (e) { }
finally {
Error.stackTraceLimit = originalStackTraceLimit;
}
})
.run({ async: false })
Based on this assumption, maybe we can find places on Node where we can safely ignore the stackTraceLimit, using this search, I found some places:
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/process/pre_execution.js#L215
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/modules/esm/worker.js#L66
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/assert.js#L263
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/repl/utils.js#L131
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/streams/compose.js#L213
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/url.js#L151
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/events.js#L501
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/encoding.js#L509
- https://github.com/nodejs/node/blob/0899bee48c64c4c2c5e2827a1ae5524a08542741/lib/internal/modules/esm/translators.js#L74
- https://github.com/nodejs/node/blob/6557c1c9b1206a85bb7d8e7450e8c3a4cff7c84b/lib/internal/worker/io.js#L277
- https://github.com/nodejs/node/blob/6557c1c9b1206a85bb7d8e7450e8c3a4cff7c84b/lib/internal/util.js#L375
- https://github.com/nodejs/node/blob/6557c1c9b1206a85bb7d8e7450e8c3a4cff7c84b/lib/querystring.js#L479
About this issue
- Original URL
- State: open
- Created 8 months ago
- Comments: 15 (14 by maintainers)
Agreed, the point here is that for
try { ... } catch { ... }
s where we do not use the error, we do not need to waste time generating the stack trace for them (afaict).