Re: CNI changes (Was: Binary Compatibility)

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > Andrew> Besides, I have some vague memory that Java semantics actually
 > Andrew> require storing into an int field to be atomic.  [Looks it
 > Andrew> up.]  Oh yes, according to Section 8.3 of the VMspec that
 > Andrew> seems to be the case.  But a method pointer might on some
 > Andrew> machines be a long rather than an int, so synchronization may
 > Andrew> be required.  I get it now, I think.
 > 
 > I've never even been certain that we correctly implement the java
 > rules for atomicity of stores.  These rules are a requirement on us to
 > generate the correct code...

That's right.  So why are we making so much fuss about this particular
instance?  It's only going to be an issue on a machine that has
multi-word function pointers.

 > Changing the lookup code will make things a little worse.  We'll need
 > an extra function call and a synchronization for every JNI method.  It
 > would be nice not to do this if possible.  This is an area I really
 > don't know much about.  Can we tell gcc that we need a certain store
 > to be atomic?

This is all rather system dependent AFAIK.

Andrew.