binary compatibility ABI (was: Re: [boehm-gc] Import 6.3 alpha 1)

Wed Aug 13 18:34:00 GMT 2003

On 13 Aug 2003, Tom Tromey wrote:
> I think the ABI implemented by -fno-assume-compiled should just be
> eliminated. It has never fully worked, and based on getting eclipse
> to work with it, I can say that it is pretty fragile.

Yeah, that's what I meant by "incomplete". If I had continued working on
it, I would've liked to fix that, perhaps by resolving method/field
references through the constant pool as we do in the interpreter.
(I never thought of -fno-assume-compiled as an "ABI" frankly. An ABI is
something that is designed, I hope, not something that happens by
accident?)
> (This happened to me a couple times, when I rebuilt some plugin with
> jikes or javac instead of ecj... ouch.)

javac from JDK 1.2 and later reorders fields and methods. I've been
bitten many times.
Jeff