add TypeInfo to GC calls

[rainers]

This is a small but inevitable step for precise scanning of GC managed objects.
It currently only passes TypeInfo for newed class and struct objects to the GC, ignoring it within the GC.

More changes are needed for dynamic and associative arrays.

If we are changing the GC API anyway, how about using an interface instead of the proxy object? That would remove a null-check and a lot of confusion.

[etcimon]

Is the m_RTInfo currently available for structs and classes? I'm curious to see an example of what it looks like before I can study this part of the upcoming GC updates on precise scanning..

[rainers]

Is the m_RTInfo currently available for structs and classes?

Yes, if it is not hit by compiler bugs. See dlang/dmd#2480

I'm curious to see an example of what it looks like before I can study this part of the upcoming GC updates on precise scanning..

Here is the precise GC as presented at last years dconf: https://github.com/rainers/druntime/tree/gcx_precise2
RTInfo generation is mostly in https://github.com/rainers/druntime/blob/gcx_precise2/src/gc/gctemplates.d

[etcimon]

Yes, if it is not hit by compiler bugs.

What is currently keeping it back from being in druntime master? Do you have a todo list?

[rainers]

Thanks for merging.

[rainers]

What is currently keeping it back from being in druntime master? Do you have a todo list?

IMO the main drawback is that it slows down compilation a bit. There are other concerns regarding the implementation, see the discussion here: http://forum.dlang.org/thread/51C49D27.8070809@gmx.de

#2 from the list has now been merged ;-)