Accept multiple --ignore-paths

[fisker]

Description

Fixes #8048

Checklist

• I’ve added tests to confirm my change works.
• (If changing the API or CLI) I’ve documented the changes I’ve made (in the docs/ directory).
• (If the change is user-facing) I’ve added my changes to changelog_unreleased/*/XXXX.md file following changelog_unreleased/TEMPLATE.md.
• I’ve read the contributing guidelines.

✨Try the playground for this PR✨

[fisker]

@prettier/core Please review.

[kachkaev]

What if we first read all files in a loop, combine them into a sing big string and then call createIgnore({ allowRelativePaths: true }).add(combinedContent)? That's how I was imagining multiple ignore paths to work from a user perspective. Combining file contents can simplify how withNodeModules is implemented: it will be just a new line appended to combinedContent, so need for introducing [undefined] etc.

Here is an example setup setup where a combined approach kinda matters: 🐸 njt / .prettierignore. This file has three sections, the first of which is specific to Prettier and the other two are copy-pasted from other files. Keeping these secions in sync between files is painful but necessary right now.

With this PR in, I can imagine removing sections named Same as in .gitignore and Shared between linters. To do that, I will be able to add this line to .prettierrc.cjs: ignorePath: [".prettierignore", ".gitignore", ".sharedlintignore"]. That will be great actually!

If we have several independent createIgnore calls plus isIgnoredFunctions.some(...), the overall behavior of the ignorer may differ from what we'd get by ‘concatenating‘ ignore files. This may mismatch user expectations and lead to confusion.

Having said that, I’m not an expert is how ignore files are usually combined. If what you have implemented is common for other linters, I might be the only person who would be confused by the output of isIgnoredFunctions.some(...), compared to having a single createIgnore.

[fisker]

1, they can be in different directories, they can't be simply concated. We need to know what the patterns related to.

#.ignore1
foo

#dir/.ignore2
bar

2. In this case

#.ignore1
foo/a.js

#.ignore2
!foo/*
foo/b.js

should foo/a.js be ignored? I think user may expect it be ignored, because it's ignored in .ignore1 file. But if we treat this case like

#.ignore
foo/a.js
!foo/*
foo/b.js

I'm not sure what's expected.

[kachkaev]

Oh I see... I forgot about ignore files sitting in different directories 🤔 You are right then, we need multiple calls to createIgnore! In that case, the full list of ignored files is a union of ignore statements in individual files. Makes sense now, ignore me 😅

[y-nk]

@fisker sorry to ask so late ; it is not told in the docs if the multiple ignore paths would respect their own root folder. for example, how would the final ignore list look like for prettier --ignore-path .ignore1,subfolder/.ignore2 --check subfolder/? (usecase is: running prettier at root level of a monorepo)

[kachkaev]

I guess we need to tweak type definitions too:

prettier/src/index.d.ts

Line 788 in e4349ca

ignorePath?: string | undefined;

↓

  ignorePath?: string | string[] | undefined;

Apart from that, LGTM! The change looks backwards-compatible, so it’d be nice to see it in v3.0 or 3.1!

If anyone else has time to take a look at the diff, please do 🙏

[lostinpatterns]

@fisker any chance of cutting a new release now that this is merged?

[kachkaev]

@lostinpatterns this change will be included into v3.0

[erikpukinskis]

@lostinpatterns FYI this is in prettier@next, it's just not tagged on Github!