Git-ignore rules inside a project should not match parts of the absolute path outside the project

[DaAitch]

• Knip: 5.19.0
• Discord discussion reference (please don't be confused about the symptoms. In the end the problem was found.)

Knip (5.19.0) may behave differently on different environments, depending on

• gitignore rules
• the absolute path, where the project is checked out

During the block

knip/packages/knip/src/util/globby.ts

Line 136 in 7c9b645

if (options.gitignore) {

git ignore rules are translated to glob ignore rules. Rules are prefixed by ** which leads to the problem, that git ignore rules now may match outside the project.

Example:

# .gitignore
# ignore all files in builds folders in the project
builds/

// bug in knip

// doesn't match any file, because absolute path matches a gitignore rule
fastGlob(['src/index.ts'], {
  cwd: '/Users/xxx/builds/project',
  ignore: ['**/builds/**', '**/builds'] // generated from git-ignore
})

// maybe-fix in knip

fastGlob(['src/index.ts'], {
  cwd: '/Users/xxx/builds/project',
  ignore: ['/Users/xxx/builds/project/**/builds/**', '/Users/xxx/builds/project/**/builds'] // fix
})

[webpro]

Thanks for the bug report.

I wonder how it's possible the **/build pattern matching files outside the pattern influences the results? Any chance a small explanation or even reproduction could be set up?

That might be a reason to either improve Knip's docs and/or implementation: perhaps by extending or doing something similar to the source mapping (dist/index.js → src/index.ts) that was recently introduced.

[DaAitch]

@webpro here it is: https://github.com/DaAitch/knip-issue-678

I simulated an absolute path, by checking in some folders.

The git-ignore rule builds/ in <PROJECT_ROOT>/Users/knipuser/cicd/builds/project42/.gitignore is translated to a glob-ignore for **/builds/**.

[webpro]

Thanks, that helped a lot.

I think I found the issue, it seems to originate here in fast-glob: mrmlnc/fast-glob#441

Need to figure out whether it can/will be fixed downstream, or work around it in Knip.

[DaAitch]

@webpro Okay interesting. Here is a red knip unit test DaAitch@5989d4a to reproduce

[webpro]

Thanks, I'd prefer fixes and test coverage in the downstream lib, let's await that first: mrmlnc/fast-glob#445