Allow skipping symlink patterns, to avoid raising a fault

[sanmai-NL]

Description

In #14627, Prettier was redesigned to ‘no longer follow symbolic links’. However, the implementation went further and turned into a breaking change. Rather than adhering to the exit status contract of only returning a non-zero exit status when a file is not formatted or when something is wrong with Prettier itself, #14627 made Prettier return a non-zero exit status upon any pattern turning out to be a pointing to symlink.

This PR fixes the original PR by @andersk to behave in line with its documentation.

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✨

[sanmai-NL]

@fisker Can you approve to kick off CI?

[fisker]

I think the behavior is expected, but maybe we can skip it if --no-error-on-unmatched-pattern is used.

[sanmai-NL]

@fisker, Can you comment on the fact that the behavior does not match the docs and was different up to the breaking change? These are things I explained in my OP.

[sanmai-NL]

@fisker This a reminder. My team is waiting for this.

[fisker]

If I remember correctly, it only throws when the pattern is used as arguments.

Say you have a/b is a symblink. The error only throws when you use prettier a/b or prettier a/b someother-pattern. It doesn't throws if you use prettier a or prettier "a/**/*". So, why are you using it as argument?

Correct me if I'm wrong.

[sanmai-NL]

This was clarified in #14627 (comment).

A tool that uses e.g., Git index changes to trigger Prettier cannot filter for symlinks in a cross-OS compatible manner.

[fisker]

Put it behind --no-error-on-unmatched-pattern then. #15533 (comment)

[sanmai-NL]

What do you mean with ‘it’? Do you mean changing this PR to fix --no-error-on-unmatched-pattern or just using --no-error-on-unmatched-pattern in the first place?

This invocation throws a fault - we have been using that parameter all along. This is also discussed in the PR, that you wanted me to open a new issue for.

│ [lefthook] executing: prettier --list-different --no-error-on-unmatched-pattern --ignore-unknown -- .lefthook.yml .mega-linter.yml .sourcery.yaml .v8rrc.yml
┃  prettier ❯ 

[error] Explicitly specified pattern ".lefthook.yml" is a symbolic link.
[error] Explicitly specified pattern ".mega-linter.yml" is a symbolic link.
[error] Explicitly specified pattern ".sourcery.yaml" is a symbolic link.
[error] Explicitly specified pattern ".v8rrc.yml" is a symbolic link.

[fisker]

I think the behavior is expected, but maybe we can skip it if --no-error-on-unmatched-pattern is used.

NOT THROW IF --no-error-on-unmatched-pattern.

[sosukesuzuki]

@fisker Please allow me to ask a question for clarification. Is what you are meaning following?

It is expected behavior that Prettier throws an error when you directly specify the symbolic link path.
However, we can accept pull requests that change that it does not throw errors when --no-error-on-unmatched-pattern is enabled.

[fisker]

Yes, that's what I expect, what do you think?

[sosukesuzuki]

I'm 👍 on your opinion.

[sanmai-NL]

@sosukesuzuki I implemented the requested change. Can you review please?

[fisker]

@sanmai-NL @andersk Please take a look if you are fine with the changes.

[sanmai-NL]

I'm fine with them too.

[sanmai-NL]

Should @sosukesuzuki still review this? Hopefully users like in #15723 can be helped out soon. The PR is very small and @fisker made some fixes and improvements already.

[junaga]

1. resolve all symlinks
2. add an option for no symlink resolving

you are not going to patch every existing tool in the world into prettier lmao.

[sosukesuzuki]

Sorry for delay my review, LGTM