[cfe-dev] TypeLoc SourceRange for builtin types

John McCall via cfe-dev cfe-dev at lists.llvm.org
Wed Sep 27 22:41:54 PDT 2017


> On Sep 28, 2017, at 1:01 AM, Erik Schmidt via cfe-dev <cfe-dev at lists.llvm.org> wrote:
> Greetings,
> 
> for getting the written source of the type of a field, I'm using
> 
> 
> pFieldDecl->getTypeSourceInfo()->getTypeLoc().getSourceRange();
> 
> 
> This works great, it returns "int" for a field like "int foo;" or even "vector<int>" for a field like "vector<int> bar;", as expected. However, for built-in types with a space inside, like "unsigned int" or "signed char", this doesn't work, strangely. For example, for a field like this:
> 
> signed char foo;
> 
> the source range returned only contains the first word, namely "signed". But it should contain "signed char", shouldn't it?
> 
> It's the same for all built-in type combinations and happens everywhere, at parameters, return types of functions and others. Am I misunderstanding something here? Can't imagine that this is a bug, since everything else works so nicely. Any hint of what I am doing wrong?

We don't currently keep absolutely all of the source information in the TypeLoc.  For example, we only store a single location in BuiltinTypeLoc.  Another example is that we don't store location information for basic CVR qualifiers.  I think that it's quite arguable that we should do these things, but it's not a simple matter because we'd want to ensure that we didn't waste a ton of memory doing things like, say, storing an empty SourceLocation for a volatile qualifier that isn't there.  And there are aspects of the current TypeLoc / TypeSourceInfo design that we'd need to reconsider in order to really do that well, like the fact that the buffer size of a TypeLoc can be determined entirely from the QualType.  Still, if someone were interested in addressing these problems, I think patches would be welcome.

John.



More information about the cfe-dev mailing list