[BUG] Paying performance costs of calling HostedGit.fromUrl with non-URLs

[Thristhart]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

When profiling npm ci, I observed that 20% of the CPU time spent in my monorepo was URL.parse, and almost all of that time is in HostedGit.fromUrl. Which is a cost that feels bad to pay in my monorepo, which has a single git URL as a specifier out of thousands of dependencies otherwise.

Expected Behavior

HostedGit.fromUrl would only be invoked if the specifier being resolved looks like a URL.

Steps To Reproduce

• Attach a profiler to an npm ci invocation, e.g. by running node --inspect-brk <path to npm.js> ci.
• Record a performance profile while npm ci is run
• URL.parse has the most self time

Environment

• npm: 11.6.0
• Node: 22.17.0
• OS: Windows 11

[Thristhart]

I'm not aware of the nuances of what types of strings should be passed to hosted-git-info, but I found locally that modifying resolve() to only call HostedGit.fromUrl if the URL matches the isURL regex worked correctly for me and dropped URL parsing from 20% to 3% of the CPU time.

e.g.

  const specIsUrl = spec && isURL.test(spec);

  const hosted = specIsUrl ? HostedGit.fromUrl(spec, {
    noGitPlus: true,
    noCommittish: true,
  }) : null;