Re: RFC: GC allocating everything with a type, part 1

[Daniel Berlin <dberlin at dberlin dot org>]

> > Or pass the type code for GC allocated data structs.  Or perhaps it's
> > not even necessary, I'm not sure yet.  The whole GC system is pretty
> > complex.
>
> 'pass the type code' how?
As a parameter (or multiple parameters for multiple different types) to 
splay_tree_new_ggc or whatever.


> So, you wouldn't need this size field any more, saving you 4 or 8
> bytes per object.

If you mean the alloc_chunk's size field (it's hard to tell, i'm not 
ready incredibly carefully though), the size fields is 15 bits + one 
bit to say if the object is large or not (large objects have the size 
field multiplied by a constant).

I don't think it actually makes any difference in the size of the alloc 
chunk structure due to alignment.

> I think you missed the '*' in the line above.  This is not a pointer
> to a struct rtx_def, it is a pointer to a separately-allocated chunk
> of memory that contains pointers to struct rtx_def.  What type does
> the separately-allocated chunk have?
None,ATM, since we don't know how to mark it separately anyway (it's 
marked as part of emit_status's marker).
Thus, we just wouldn't cut off marking there.