Change TypeInfo.next to inout function

[9rnsr]

In 2.060, following code has changed to not to work.

TypeInfo ti = typeid(SomeType);
while (ti.next) { ti = ti.next; }
// ti.next now returns const(TypeInfo), then rebinding is rejected.

This change fixes the problem.

[alexrp]

Fails on Darwin32 due to a compiler bug...

[MartinNowak]

Locate TypeInfo_Vector after TypeInfo_AssociativeArray

Why?

[MartinNowak]

The OSX32 bug happens when compiling rt.lifetime and I've reduced it to this.

cat > bug.d << CODE
void bug()
{
    auto sizeelem = typeid(int).next.tsize;
}
CODE

dmd -c -m32 -O -release -inline bug.d

[MartinNowak]

OK, it's an unrelated bug Bugzilla 9096.

[alexrp]

Does this still fail on OS X?

[MartinNowak]

Does this still fail on OS X?

Yes, the problem is related to the inout addition.
A workaround was to simply use the following.

auto ti = typeid(int).next;
auto sizeelem = ti.tsize;

But accessing a TypeInfo field of a return type or an array element is pretty common, so merging this pull without fixing Bugzilla 9096 first might break a lot of code.

[9rnsr]

By removing workaround for bug 9231, the symptom of bug 9096 is now disappeared. But I don't know why that...

[alexrp]

Well... I suppose we can merge this then.

[andralex]

@9rnsr: if we have a passing unittest for bug 9096, then you may want to close it as resolved.

[9rnsr]

@andralex Sorry, I don't know whether the bug 9096 is resolved. It had fortunately just gone from the test result for this pull.