Parser: avoid skipping inactive code if we are building syntax trees.

[nkcsgexi]

If syntax trees are requested, we shouldn't skip inactive code. Notice the
inactive code won't be skipped in SwiftSyntax because we always set
PerformConditionEvaluation false for the in-process parser.

This is mostly needed for testing purposes where we add -verify-syntax-tree
to regular compiler invocations.

rdar://50837165

[nkcsgexi]

@swift-ci please smoke test

[compnerd]

Am I mistaken, or does this really not actually verify anything? It would require that the user manually verify that the inactive region is processed.

[nkcsgexi]

This verifies we can get syntax nodes for inactive code like print("wat"), otherwise an error message will be emitted to complain about unknown syntax nodes.

[akyrtzi]

How about we add an extra step to sanity check that everything is parsed and round-tripped ?
E.g. add a RUN line to check that printing the syntax-tree will result in the same output as the input source file.
This will ensure that everything is parsed as expected.

Mistake

[rintaro]

Can we disable diagnostics using DiagnosticSuppression?

[nkcsgexi]

Good point!

[nkcsgexi]

@swift-ci please smoke test

[nkcsgexi]

@akyrtzi @rintaro Two additional commits were added to disable diagnostics in the inactive code though we parse inside the inactive code. Notice the existing design of syntax tree verifier works as a post-processor. To make it compatible with DiagnosticSuppression, we have to refactor the verifier to as on-the-fly.

[jrose-apple]

I'm a little concerned about this since we just saw issues with #if canImport(…) getting evaluated in inactive blocks. What keeps something like that from happening here?

[akyrtzi]

I think that it shouldn't be the responsibility of SyntaxTreeCreator to emit verification diagnostics, I preferred it when the SyntaxVerifier was a separated, self-contained, functionality.

What do you think about an alternative solution; we could just disable the syntax-tree verifier in the case where #if swift showed up, which is pretty uncommon.
Instead of:

} else if (SyntaxContext->isEnabled()) {
  // We shouldn't skip code if we are building syntax tree.
  // The parser will keep running and we just discard the AST part.
  DiagnosticSuppression suppression(Context.Diags);
  SmallVector<ASTNode, 16> dropedElements;
  parseElements(dropedElements, false);
} else {

we'd do:

} else if (SyntaxContext->isEnabled()) {
  // Disable the syntax tree verifier, we are skipping parsing, so 'unknown' node is expected.
  <indicate that syntax tree verifier must not run>
} else {

This has these benefits:

• Preserves the existing behavior, which is that no parsing should occur
• Much simpler.

[nkcsgexi]

@swift-ci please smoke test

[nkcsgexi]

@swift-ci please smoke test

[nkcsgexi]

@swift-ci Please clean test Linux platform

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 7af15fab462f65fafe1929455570662d99b13f6a

[nkcsgexi]

@swift-ci Please smoke test OS X platform