manually inlining validator functions improves performance

Lets take the isAbsolute function of posix in the path module.

  /**
   * @param {string} path
   * @returns {boolean}
   */
  isAbsolute(path) {
    validateString(path, 'path');
    return path.length > 0 &&
           StringPrototypeCharCodeAt(path, 0) === CHAR_FORWARD_SLASH;
  },

Benchmark it:

./node benchmark/path/isAbsolute-posix.js 
path/isAbsolute-posix.js n=100000 path="": 26,199,553.088023424
path/isAbsolute-posix.js n=100000 path=".": 19,512,560.625525866
path/isAbsolute-posix.js n=100000 path="/foo/bar": 15,003,530.33068681
path/isAbsolute-posix.js n=100000 path="/baz/..": 20,841,496.252698973
path/isAbsolute-posix.js n=100000 path="bar/baz": 25,044,717.342815597

inline validateString

  /**
   * @param {string} path
   * @returns {boolean}
   */
  isAbsolute(path) {
    if (typeof path !== 'string')
      throw new ERR_INVALID_ARG_TYPE('path', 'string', path);
    return path.length > 0 &&
           StringPrototypeCharCodeAt(path, 0) === CHAR_FORWARD_SLASH;
  },

./node benchmark/path/isAbsolute-posix.js 
path/isAbsolute-posix.js n=100000 path="": 49,143,043.605605446
path/isAbsolute-posix.js n=100000 path=".": 36,665,695.025748484
path/isAbsolute-posix.js n=100000 path="/foo/bar": 20,950,367.322790273
path/isAbsolute-posix.js n=100000 path="/baz/..": 22,806,405.772027623
path/isAbsolute-posix.js n=100000 path="bar/baz": 41,722,470.470921524

Annoying...

can you share the benchmark?

This is not surprising, the inline version doesn't need hideStackFrames.
The only other solution I see is using a macro system (that was removed some years ago).

Maybe we could make the "stack hiding" feature opt-in/opt-out. I'm not sure what's the use-case behind that, but I think many folks would gladly take the extra stack lines as a tradeoff for the increase performance.

What is going here? V8 ought to be able to inline the function during the JIT compilation, no?

function validateString(value, name) {
  if (typeof value !== 'string') {
    throw new ERR_INVALID_ARG_TYPE(name, 'String', value);
  }
}

Why won't it?

I dont know but I think this is because you pass by val and not by ref.

Maybe we could make the "stack hiding" feature opt-in/opt-out.

Was anything decided on this?

Navigation Menu

Repository navigation

manually inlining validator functions improves performance #154

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

manually inlining validator functions improves performance #154

Description

Activity

ronag commented on Mar 19, 2024

targos commented on Mar 19, 2024

aduh95 commented on Mar 19, 2024

lemire commented on Mar 19, 2024

Uzlopak commented on Apr 8, 2024

Dhruv-Garg79 commented on Jan 16, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions