Update php-parser, fix unary snapshot tests, and make multiple fixes for attribute formatting.

[acharron-hl]

Fixes #2293
Fixes #2486

Dependency update note

This PR relies on a new version of php-parser: 3.7.0.

I was initially setting out to just fix the attribute formatting issues, but that required a new version of the php-parser. Unfortunately the currently published 3.6 has a regression around the unary operator reported as #2486 which means we can't just update to that.

Attribute formatting fixes

Fixes #2293

The updated php-parser exposes attribute groups as distinct AST nodes on methods, functions, class constants, and enum cases. I madechanges so attributes and their comments format correctly.

Changes

• getCommentChildNodes — attribute groups are now comment attachment targets on function/method, classconstant, and enumcase, so comments between #[...] blocks attach to the right node instead of being skipped or misplaced.
• printAttrs / printAttrGroup — each attribute group is printed through printAllComments(), so leading and trailing comments on groups are actually emitted.
• enumcase printing — attributes on enum cases are now printed before case (new coverage in tests/enum/enum-case-attributes.php).
• handleClassMemberStatementComments — comments before a property or constant identifier are attached to the enclosing member statement, keeping modifiers like public static together when comments appear between modifiers and the name.

Snapshot impact

Test	What changed
tests/attributes/attributes.php	Comments between #[S] and #[T] now print in place; //Testing T no longer incorrectly trails the previous method's }.
tests/comments/method.php	Dangling comment in empty () stays inside parens; inline comments between static and $foo are all preserved.
tests/enum/enum-case-attributes.php	New snapshot for attributed enum cases.

---

Unary operator / exponentiation fix (upstream parser)

Fixes #2486

This change comes from php-parser, not from new formatting logic in the plugin. PHP gives ** higher precedence than unary + and -, so unparenthesized expressions parse differently from explicitly parenthesized ones:

+$a ** 1   // PHP parses as:  +($a ** 1)
(+$a) ** 1 // PHP parses as:  (+$a) ** 1

The old parser incorrectly parsed +$var ** 2 as (+$var) ** 2. The formatter faithfully reproduced that wrong AST, so both forms printed identically. The updated parser builds the correct tree — unary wrapping exponentiation for the unparenthesized form, and a bin(**) with parenthesizedExpression: true when parens are explicit.

Only the snapshots needed updating.

Snapshot impact

Test	Input	Old output	New output
tests/parens/bin.php	$var = +$var ** 2;	(+$var) ** 2	+($var ** 2)
tests/parens/unary.php	$a = +$a ** 1;	(+$a) ** 1	+($a ** 1)

Explicitly parenthesized inputs such as (+$var) ** 2 and (+$a) ** 1 are unchanged.

This is a semantic correction: the old output did not match how PHP actually parses unparenthesized +$x ** n expressions.

[jorgsowa]

Please don't merge fixes and put into description only relevant information. It's hard to review such PRs.

This PR relies on a new / unpublished version of php-parser (which will likely be version 3.6.1 or 3.7.0.

This is not acceptable solution.

[acharron-hl]

This is not acceptable solution.

I'm aware. That's why I noted in the first section

It should not be merged until that version is published. In the meantime, this PR depends on the commit at the tip of the main branch over there (which includes the unary operator fix.

I was initially setting out to just fix the attribute formatting issues, but that required a new version of the php-parser. Unfortunately the currently published 3.6 has a regression around the unary operator reported as #2486 which means we can't just update to that.

---

put into description only relevant information

I don't think it's necessarily overly verbose? I'm mostly just explaining which snapshots are updating any why. The unary operator change in particular is a bit finicky and not immediately obvious to me what was wrong with the initial implementation, which is why I wrote up the examples.

[czosel]

@acharron-hl php-parser v3.7.0 has just been released 👍

[acharron-hl]

@jorgsowa @czosel I've updated the php-parser version to 3.7.0. I removed the packageManager declaration that was auto-added to package.json as well when I ran yarn install with corepack enabled on my node version.

The tests passed locally for me at least.

[jorgsowa]

You can see that something is messed up with comments. Why did this comment change its position?

Do you know whether it's a fault of the PHP parser or your PR?

[acharron-hl]

This is from the changes I made to comment/attribute formatting (Item 1 in my snapshot changes listed in the PR description).

So this is the raw unformatted code

    }

    // Testing S
    #[S]
//Testing S-T
#[T] //Testing T

This is the old snapshot

    } //Testing T          ← comment moved here (wrong)
    // Testing S
    //Testing S-T          ← moved above #[S] (wrong)
    #[S]
    #[T]   

And my udpated snapshot

    }                       ← no stray comment
    // Testing S
    #[S]
    //Testing S-T          ← stays between #[S] and #[T]
    #[T] //Testing T        ← comment stays with #[T]
    private function u()

It's kind of hard to tell in the git diff that's for sure. One of the negatives of such a huge snapshot is that the input/output are hundreds of lines apart when looking at the snapshot.

[acharron-hl]

@jorgsowa So it seems we're held up on if the project should preserve comments in between visibility modifiers or not. If you want I can implement it. Anything to get our growing list of enums out of our .prettierignore file.

[jorgsowa]

Sorry, for delay. Looks good. Thanks for contribution