diff options
| author | Faisal Vali <faisalv@yahoo.com> | 2017-05-02 20:56:34 +0000 |
|---|---|---|
| committer | Faisal Vali <faisalv@yahoo.com> | 2017-05-02 20:56:34 +0000 |
| commit | 999f27e3739efbe43edd1baa09dedd9dcea68d6f (patch) | |
| tree | a8917fe0ceede6935aa02b2275b47826e6abab0e /clang/lib | |
| parent | c7695a8e4565afb256aad1a705ef292bc3625a36 (diff) | |
| download | bcm5719-llvm-999f27e3739efbe43edd1baa09dedd9dcea68d6f.tar.gz bcm5719-llvm-999f27e3739efbe43edd1baa09dedd9dcea68d6f.zip | |
Fix PR32831 (Try Again): 'this' capture while instantiating generic lambda call operator specialization
When computing the appropriate cv-qualifiers for the 'this' capture, we have to examine each enclosing lambda - but when using the FunctionScopeInfo stack we have to ensure that the lambda below (outer) is the decl-context of the closure-class of the current lambda.
https://bugs.llvm.org/show_bug.cgi?id=32831
This patch was initially committed here: https://reviews.llvm.org/rL301735
Then reverted here: https://reviews.llvm.org/rL301916
The issue with the original patch was a failure to check that the closure type has been created within the LambdaScopeInfo before querying its DeclContext - instead of just assuming it has (silly!). A reduced example such as this highlights the problem:
struct X {
int data;
auto foo() { return [] { return [] -> decltype(data) { return 0; }; }; }
};
When 'data' within decltype(data) tries to determine the type of 'this', none of the LambdaScopeInfo's have their closure types created at that point.
llvm-svn: 301972
Diffstat (limited to 'clang/lib')
| -rw-r--r-- | clang/lib/Sema/SemaExprCXX.cpp | 53 |
1 files changed, 39 insertions, 14 deletions
diff --git a/clang/lib/Sema/SemaExprCXX.cpp b/clang/lib/Sema/SemaExprCXX.cpp index d65570fcef7..9b88cddbc96 100644 --- a/clang/lib/Sema/SemaExprCXX.cpp +++ b/clang/lib/Sema/SemaExprCXX.cpp @@ -901,17 +901,36 @@ static QualType adjustCVQualifiersForCXXThisWithinLambda( // capturing lamdbda's call operator. // - // The issue is that we cannot rely entirely on the FunctionScopeInfo stack - // since ScopeInfos are pushed on during parsing and treetransforming. But - // since a generic lambda's call operator can be instantiated anywhere (even - // end of the TU) we need to be able to examine its enclosing lambdas and so - // we use the DeclContext to get a hold of the closure-class and query it for - // capture information. The reason we don't just resort to always using the - // DeclContext chain is that it is only mature for lambda expressions - // enclosing generic lambda's call operators that are being instantiated. - + // Since the FunctionScopeInfo stack is representative of the lexical + // nesting of the lambda expressions during initial parsing (and is the best + // place for querying information about captures about lambdas that are + // partially processed) and perhaps during instantiation of function templates + // that contain lambda expressions that need to be transformed BUT not + // necessarily during instantiation of a nested generic lambda's function call + // operator (which might even be instantiated at the end of the TU) - at which + // time the DeclContext tree is mature enough to query capture information + // reliably - we use a two pronged approach to walk through all the lexically + // enclosing lambda expressions: + // + // 1) Climb down the FunctionScopeInfo stack as long as each item represents + // a Lambda (i.e. LambdaScopeInfo) AND each LSI's 'closure-type' is lexically + // enclosed by the call-operator of the LSI below it on the stack (while + // tracking the enclosing DC for step 2 if needed). Note the topmost LSI on + // the stack represents the innermost lambda. + // + // 2) If we run out of enclosing LSI's, check if the enclosing DeclContext + // represents a lambda's call operator. If it does, we must be instantiating + // a generic lambda's call operator (represented by the Current LSI, and + // should be the only scenario where an inconsistency between the LSI and the + // DeclContext should occur), so climb out the DeclContexts if they + // represent lambdas, while querying the corresponding closure types + // regarding capture information. + + // 1) Climb down the function scope info stack. for (int I = FunctionScopes.size(); - I-- && isa<LambdaScopeInfo>(FunctionScopes[I]); + I-- && isa<LambdaScopeInfo>(FunctionScopes[I]) && + (!CurLSI || !CurLSI->Lambda || CurLSI->Lambda->getDeclContext() == + cast<LambdaScopeInfo>(FunctionScopes[I])->CallOperator); CurDC = getLambdaAwareParentOfDeclContext(CurDC)) { CurLSI = cast<LambdaScopeInfo>(FunctionScopes[I]); @@ -927,11 +946,17 @@ static QualType adjustCVQualifiersForCXXThisWithinLambda( return ASTCtx.getPointerType(ClassType); } } - // We've run out of ScopeInfos but check if CurDC is a lambda (which can - // happen during instantiation of generic lambdas) + + // 2) We've run out of ScopeInfos but check if CurDC is a lambda (which can + // happen during instantiation of its nested generic lambda call operator) if (isLambdaCallOperator(CurDC)) { - assert(CurLSI); - assert(isGenericLambdaCallOperatorSpecialization(CurLSI->CallOperator)); + assert(CurLSI && "While computing 'this' capture-type for a generic " + "lambda, we must have a corresponding LambdaScopeInfo"); + assert(isGenericLambdaCallOperatorSpecialization(CurLSI->CallOperator) && + "While computing 'this' capture-type for a generic lambda, when we " + "run out of enclosing LSI's, yet the enclosing DC is a " + "lambda-call-operator we must be (i.e. Current LSI) in a generic " + "lambda call oeprator"); assert(CurDC == getLambdaAwareParentOfDeclContext(CurLSI->CallOperator)); auto IsThisCaptured = |

