over-eager verification

Wed Mar 6 22:51:00 GMT 2002

Tom Tromey wrote:
> We could get better performance doing lazy verification. But then we
> wouldn't be making a compatible Java implementation. We'd be making
> something that is almost but not quite Java.

Well, that is unavoidable, since Java is defined by Sun's
implementation. We can be close enough for some large number of
real programs, but there will always be corner cases. It's like
implementing Wine - you're going to have to decide which features
and bugs you want to be 100% ompatible, and in which cases we will
be different.
> I think that compatibility is important. I prefer compatibility and
> adherence to the specification (to the extent possible, given its
> occasional opacity) over divergence, even when the divergence has
> benefits.

In general, I tend to agree. However ...
Gcc traditionally has uses flags like -ansi and -pedantic and
POSIXLY_CORRECT to control standards-compliance. It is very desirable
that GNU software provide a way to be strictly compatible with
published standards or popular third-party implementations. However,
GNU does not always automatically follow the strict letter of a standard
or other implementation, when there is a better GNU mechanism.
Whether eager vs lazy verification is a feature where we should
pedantically follow the specification is a matter one can disagree on.
Adherence to the specification is a good reason to do it one way - but
it does not automatically overrule other reasons for doing it a
different way.
Anyway, we can leave it the way it is for now.
-- 
	--Per Bothner
per@bothner.com http://www.bothner.com/per/