Add diagnostic for label with string segment

[kimdv]

Resolves #1831

[kimdv]

@swift-ci please test

[kimdv]

@ahoppen testDiagnoseAvailability18 fails.

It fails formatting the same code, this is the result

@available(*, unavailable, message: """
  foobar message.
  
""") func multilineMessage() {
}
multilineMessage()

A quick debug I think it's because we do some post processing of multiline string when parsing a RawStringLiteralExprSyntax.

We can use the same method, but that require we convert the RawStringSegmentSyntax to a RawStringLiteralSegmentsSyntax.Element and then back to RawStringSegmentSyntax.

Or we should try to extract the postProcessMultilineStringLiteral into multiple functions so we better can reuse it.

What do you think?

[ahoppen]

I think we should try just re-using postProcessMultilineStringLiteral. Mapping between RawStringLiteralExprSyntax and RawStringLiteralSegmentsSyntax.Element doesn’t result in an allocation on the bump allocator, so we shouldn’t be leaking any memory that way, so the mapping should be fine.

[kimdv]

@ahoppen I was just debugging and trying to figure out why we get this error

I've tried to debug, but I can't find the reason

Trees do not match due to token in:
stringSegment("  foobar message.") [2:1...2:18]
when expecting:
stringSegment("foobar message.") [2:3...2:18]

[ahoppen]

As the failure message says: What we do is remove all the trivia of the tree, run BasicFormat and check that we end up with the same tree. This basically tests BasicFormat to check that we indeed separate different tokens by space.

But when you remove the leading spaces from the closing """, you are really changing the semantic meaning of the string literal because the two spaces in front of foobar message. now become part of the string literal.

That’s why in SwiftParserTest/Assertions.swift:735 (inside TriviaRemover, we check if the token is inside a StringLiteralSyntax, in which case we just don’t remove trivia from inside this subtree. You just need to add SimpleStringLiteralExprSyntax here.

And as I’m looking at the code, I’m realizing that we could actually simplify TriviaRemover

/// Removes trivia from all tokens that don’t occur inside multiline string
/// literals.
///
/// We keep trivia inside multiline string literals because the indentation of
/// the closing quote of a multi-line string literals has impact on how much
/// leading trivia is stripped from the literal’s content.
class TriviaRemover: SyntaxRewriter {
  override func visit(_ node: StringLiteralExprSyntax) -> ExprSyntax {
    if node.openingQuote == .multilineStringQuoteToken() {
      return ExprSyntax(node)
    } else {
      return super.visit(node)
    }
  }

  override func visit(_ node: SimpleStringLiteralExprSyntax) -> ExprSyntax {
    if node.openingQuote == .multilineStringQuoteToken() {
      return ExprSyntax(node)
    } else {
      return super.visit(node)
    }
  }

  override func visit(_ token: TokenSyntax) -> TokenSyntax {
    return token.with(\.leadingTrivia, []).with(\.trailingTrivia, [])
  }
}

[kimdv]

@swift-ci please test

[kimdv]

@swift-ci please test

[kimdv]

@swift-ci please test

[ahoppen]

@swift-ci Please test Windows

[ahoppen]

A couple comments to prevent round-trip failures, otherwise LGTM.

[kimdv]

@swift-ci Please test

[kimdv]

@swift-ci Please test

[kimdv]

@swift-ci Please test windows