[patch] Fix source range of CXXNewExprs

Clang Frontend
1

Hi,

the attached patch fixes the source range on CXXNewExprs. For code like

  foo:A* a = new foo::A;

the CXXNewExpr's source range's end would point to `foo` instead of
`A`. This was because Sema::BuildCXXNew() uses the allocated type's
TypeLoc's source range's end as end location for CXXNewExpr if no
parens are present (new foo::A() is handled correctly already), and
TypeSpecTypeLoc has only a start location.

My patch adds an end location to TypeSpecTypeLoc and fills it in in
TypeSpecLocFiller::VisitTagTypeLoc(). The rest of the patch consists
of copying this new TypeSpecTypeLoc field every time a TypeSpecTypeLoc
is copied, which happens in surprisingly many places.

Acurate sourceranges on all ast nodes are important for writing rewriters.

Nico

newrange.patch (10.6 KB)

2

This should be an ElaboratedTypeLoc. If it isn't, that's the bug, not changing
TypeSpecTypeLoc.

John.

3

It is an ElaboratedTypeLoc, but TypeLoc::getEndLoc() skips the
ElaboratedTypeLoc's position and moves on the next loc:

SourceLocation TypeLoc::getEndLoc() const {
  TypeLoc Cur = *this;
  while (true) {
    switch (Cur.getTypeLocClass()) {
    default:
      break;
    case Qualified:
    case Elaborated:
      Cur = Cur.getNextTypeLoc();
      continue;
    }
    break;
  }
  return Cur.getLocalSourceRange().getEnd();
}

Nico

4

Ah. So the problem is that the TypeSpecTypeLoc's location should be
the location of the qualified identifier, not the location of the start of the
qualification.

John.

5

Like this? If so, I guess the other calls to SetTypeSpecType() where a
annot_typename is parsed need this change, too?

Nico

sourceranges2.patch (1.32 KB)

6

On SemaType.cpp:1561, try changing
  if (Keyword == ETK_Typename) {
to
  if (DS.getTypeSpecType() == TST_typename)

If that works, please do the same thing to SemaType.cpp:1579.

John.

7

That doesn't seem to help.

Nico

8

Ah yes, I see the problem. You're right, we need to use getAnnotationEndLoc(),
both here and elsewhere. I think we probably also need the change to
SemaType.cpp, though.

John.

9

Third time's the charm? Please take a look :slight_smile:

Nico

sourceranges3.patch (4.39 KB)

10

Looks good, thanks!

John.

11

r119966. Thanks!