Post #130,140
12/10/03 8:59:26 PM
12/10/03 9:02:00 PM
|
Java FP sucks (surprise)
While under jb's cracking whip, I ran across this guy's papers:
[link|http://www.cs.berkeley.edu/~wkahan/|http://www.cs.berkeley.edu/~wkahan/]
He was one of the creators of the IEE754 floating point standard. He wrote this paper on Java floating point:
[link|http://www.cs.berkeley.edu/~wkahan/JAVAhurt.pdf|http://www.cs.berkel...ahan/JAVAhurt.pdf]
Very fun.
Abstract: Java s floating-point arithmetic is blighted by five gratuitous mistakes:
1. Linguistically legislated exact reproducibility is at best mere wishful thinking.
2. Of two traditional policies for mixed precision evaluation, Java chose the worse.
3. Infinities and NaNs unleashed without the protection of floating-point traps and flags mandated by IEEE Standards 754/854 belie Java s claim to robustness.
4. Every programmer s prospects for success are diminished by Java s refusal to grant access to capabilities built into over 95% of today's floating-point hardware.
5. Java has rejected even mildly disciplined infix operator overloading, without which extensions to arithmetic with everyday mathematical types like complex numbers, intervals, matrices, geometrical objects and arbitrarily high precision become extremely inconvenient.
To leave these mistakes uncorrected would be a tragic sixth mistake.
Tee hee hee...
Now, things may have improved, and in any case I would never even think of Java as a numerical computing platform, but perhaps these boners are indicative of a sloppy mindset in the creation of the thing.
edit: Author's bold reinstated after Acrobat cut/paste
-drl
Edited by deSitter
Dec. 10, 2003, 09:02:00 PM EST
|
Post #130,143
12/10/03 9:05:53 PM
|
Java: a solution in search of a problem. Nice find.
bcnu,
Mikem
Java, Junk. Both start with a "J", both have four letters. Coincidence? I think not.
|
Post #130,148
12/10/03 9:26:42 PM
|
It's an excellent paper!
It's not just about Java, rather, Rampant Stupidity in Computing. Java just happens to be the exemplar of that theme.
TB *needs* to read it end to end!
-drl
|
Post #130,194
12/11/03 3:58:45 AM
|
Sheesh!
A) By now,, *everyone* knows that all these machines are (nada mas que..) fucking adding machines: stuff it in the accumulator, add it to a register. A couple times.
B) Now you're telling us: the fucking software manages to screw even THAT up?
Shakespeare was right about that "sound & fury" bit.
|
Post #130,253
12/11/03 10:43:29 AM
|
Seen that
Over the summer when working on meat processing calculation engine - lots of floating point arithmetic - taking conversion percentages and such.
Java's willingness to distinguish between -0.0 and 0.0 and its happy silent conversions to NaN when doing something foolish like dividing by zero made the thing fucking hard to debug.
Java is a hack job.
I remain amazed that Smalltalk (alone? Don't know) has had arbitrary precision math since the beginning - including exact representations like fractions and such. While nearly every language produced since then has had only crude approximations built in.
"I believe that many of the systems we build today in Java would be better built in Smalltalk and Gemstone."
-- Martin Fowler, JAOO 2003
|
Post #130,349
12/11/03 2:22:31 PM
|
Well, it definitely did....(the article is from 1998)
and it accurately describes the situation back then, in particular the conflict with Sun's version and Gnu's version (gnu.math) (among others).
Not to say Java's FP doesn't (still) have issues - it does. Check out the [link|http://www.sonic.net/~jddarcy/Borneo/| Borneo ] web site.
|
Post #130,350
12/11/03 2:24:53 PM
|
Re: Well, it definitely did....(the article is from 1998)
What was perhaps most interesting was his pointing out that Sun and Gosling had a smirking, smart ass attitude about the whole thing - what, us worry? - and how Cahan lumps them with Microsoft as joint offenders.
-drl
|
Post #130,488
12/12/03 2:25:17 PM
|
Sun - the other Microsoft
"I believe that many of the systems we build today in Java would be better built in Smalltalk and Gemstone."
-- Martin Fowler, JAOO 2003
|
