Addendum about object headers.

The argument *against* having a header with the unboxed object is basically
about object layout compatibility with C. But C compatibility is about
value type layout, not reference type object layout. C doesn't *have* unboxed
objects.

Which brings me to wonder whether we aren't pushing the wrong requirement.
One option here is to partition the problem by type. Value types would have
no header, would be named by REF 'vt, and would require a long-form store
barrier in a deferred-RC scheme. Unboxed objects (if we support that)
*would* have a header, and interior references can be used directly by
means of interior references.

The case that has me worried here is regions. When the language supports
LIFO regions, there is an increasing temptation to make use of inner
pointers into lifetime-safe data structures.
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to