Re: RFC: Re: Re: [commit-cp] classpath ./ChangeLog examples/gnu/classpath/ex...

>>>>> "Andrew" == Andrew John Hughes <gnu_andrew@xxxxxxxxxxxxxx> writes:

Andrew> However, the result is useless as nothing can even handle it,
Andrew> let alone bootstrap with it.  I was thinking of taking a look
Andrew> at fixing something like jamvm to handle it, but if you're
Andrew> already working on something similar with gcjx, then this may
Andrew> be a better target (particularly as the verifier catches the
Andrew> ldc construct, while I guess that's what's causing the
Andrew> segfault I get with JamVM).  Is there anything I can do to
Andrew> help on this score?

One question I haven't looked into is how strict we need to be about
rejecting "ldc <class>" in class files with pre-1.5 versions.

On the one hand, it is probably pedantically correct to reject this.
On the other hand, it can't possible hurt type safety (or anything
else I can think of) to accept it.

I wonder what Sun does.

The reason this matters is that if we take the lax approach, getting
gij to accept this opcode (and thus, 1.5 bytecode in general,
depending on class availability) is completely trivial... I think it
is adding one check in one place in the runtime verifier.

The alternative actually is only slightly more complicated.

Anyway, if I knew the answer to this, that would help.
We definitely should be able to run 1.5 bytecode with gij well before
all of gcjx is ready... we don't need to depend on that.

Tom