Re: bug 10491 was Re: More astonishing progress in japi scores

[Casey Marshall]

>>>>> "Robert" == Robert Schuster <address@hidden> writes:

Robert> Hi
>> 
>> I've done the same comparison 6 weeks ago with Tom Tromey, and we
>> were at 75% back then.
>> 
>> 5 % in 6 weeks. 20 % to go. You do the math. :)

That might turn out to be the longest 6 months ever, since Classpath
is still missing the badly-written web browser and HTML WYSIWYG
composer (javax.swing.text.html) :)

Robert> I am excited too but on the other hand I worry about a lot
Robert> bugs popping up ...

I don't worry that much about bugs. More bugs means more people are
interested, and it means better code as those bugs are fixed.

Robert> After spending some time on ideas to fix this the following
Robert> came to my mind:
Robert> a) do the NaN,Infinity in Java and the rest in C
Robert> b) do both in C because its faster
Robert> c) provide a pure Java and a pure C version, making
Robert> JNode/... and real VMs equally happy

I am a fan of (c), and in general I like it better if there is a pure
Java version first. I also don't tend to blindly believe that a C
version will be universally faster.

Robert> c) seems to be my favorite for an umbrella project like
Robert> classpath but I dont know how I could make the two
Robert> implementations optional.  Any ideas about this and my
Robert> proposal?

I'd suggest doing a pure Java version, if it's difficult to provide
both. Classpath should be complete and correct, and that can be done
in most of the library in Java code only, and done simpler and
clearer.

-- 
Casey Marshall || address@hidden