invokespecial on abstract method

[Andrew Haley]

Erik Poupaert writes:
 > 
 > >>> I'm sure it wouldn't be hard for Sun to make their VM fully evaluate 
 > >>> all linkage before running anything, if they wanted to do that. But by 
 > >>> design, Java doesn't.
 > 
 > In terms of optimizing performance/footprints ahead-of-time, is
 > there a cost associated to late linking? I somehow guess there
 > is. More likely than not, this strategy of late linking may
 > complicate other ongoing efforts that focus on these optimizations.

Yes.  But if you really want to link statically without indirect
dispatch, you'll still be able to do that.

 > There may be enormous advantages in doing exactly the opposite of
 > what Sun does.
 > 
 > When competing with the JDK, it may exactly be this different set
 > of trade-offs, better attuned to particular situations, that could
 > attract a sizeable following to GCJ.

Maybe.  Please don't worry: I have no intention of disabling the
useful features of gcj.  Even with indirect dispatch as the default,
you'll still be able to do something like

   gcj a.java b.java c.java -o fluff.o

and all the references between a, b, and c will be direct.  Indirect
dispatch only applies to external linkage.

Andrew.