Code completion: Stop suggesting old-style protocol metatype syntax in Swift 6 mode

[AnthonyLatsis]

In Swift 6 mode, [.]Protocol¹ is no longer a valid suggestion. Per SE-0335, we should suggest two distinct [.]Type¹ completions for a plain protocol/composition — one aiming at any P.Type and another aiming at (any P).Type — that also insert a missing any prefix, and parenthesize the acted-upon existential in the second case.

Test case (should not succeed):

// RUN: %empty-directory(%t)
// RUN: %target-swift-ide-test -batch-code-completion -source-filename %s -filecheck %raw-FileCheck -completion-output-dir %t -swift-version 6

protocol P {}

let _: P#^COMPLETE1?check=META^#
let _: any P#^COMPLETE2?check=META^#
let _: (any P)#^COMPLETE3?check=META^#

// META: Keyword/None: .Protocol[#(any P).Type#]; name=Protocol

Environment

• Swift 5.9-dev (c1d5118)

Footnotes

1. [x] means x is optional ↩ ↩²

[Rajveer100]

@AnthonyLatsis
Could you give me a brief insight on this?

[AnthonyLatsis]

@Rajveer100 After updating the description, I am no longer positive this can be qualified as a beginner-friendly issue, but here is some tips:

• Code completion can be tested using the swift-ide-test executable found in the same place as swift-frontend (see test/IDE for example invocations).
• In code completion mode, this executable searches for code completion tokens spelled #^TOKEN_NAME^#.
• The places the Protocol completion is suggested are CompletionLookup::getPostfixKeywordCompletions and CompletionLookup::getTypeCompletions.

[AnthonyLatsis]

@ahoppen @rintaro Please review the good first issue status.

[ahoppen]

Yes, I agree that this could be a good first issue.

[aniket-singh-01]

Hi Sir,
Can you assign this to me as I think this could solve it.

protocol P {}
P#^COMPLETE?check=META^#
// META: Keyword/CurrNominal: .Protocol[#(any P).Type#]; name=Protocol

changes to

protocol P {}
P.#^COMPLETE?check=META^#
// META: Keyword/CurrNominal: .Type[#any P.Type#]; name=Type
// META: Keyword/CurrNominal: .Type[#(any P).Type#]; name=Type

[AnthonyLatsis]

Note that a solution should also handle the any P and (any P) cases. Only the existential metatype completion should be suggested for any P, and only the singleton metatype completion — always spelled using any syntax, not only in Swift 6 mode — should be suggested for (any P).

[aniket-singh-01]

// META: Keyword/CurrNominal: .Type[#any P.Type#]; name=Type
// META: Keyword/CurrNominal: .Type[#(any P).Type#]; name=Type
The completion suggestions include .Type[#any P.Type#], which represents the existential metatype for any P, and .Type[#(any P).Type#], which represents the singleton metatype for (any P).

isn't this correct Sir?

[AnthonyLatsis]

Your completion suggestions are fine for P#^COMPLETE^#, but not for any P#^COMPLETE^# or (any P)#^COMPLETE^#. Only one .Type completion is reasonable for those two. They should be addressed because they also produce a .Protocol completion suggestion:

• any P.Protocol is not a sensible type at all.
• (any P).Protocol is valid pre-Swift 6 but a mixture of the old and new syntaxes, which I believe we do not want to suggest.

[aniket-singh-01]

any P.#^ANY_COMPLETE?check=META^#
// META: Keyword/CurrNominal: .Type[#any P.Type#]; name=Type

(any P).#^PAREN_COMPLETE?check=META^#
// No completion suggestion

When we trigger code completion at any P.#^ANY_COMPLETE?check=META^#, the completion suggestion is .Type[#any P.Type#], which represents the existential metatype for any P.

There is no completion suggestion provided for (any P)#^COMPLETE?check=META^# because any P.Protocol is not a sensible type and (any P).Protocol is a mixture of the old and new syntaxe

I hope it works.