Re: [xml-dev] More predictions to mull over

[noah_mendelsohn@us.ibm.com]

Not sure whether it's still true, but 10+ years ago a lot of scientists 
were still using FORTRAN in part because the optimizations people were 
putting into the compilers were more suited to numeric codes, while the 
optimizations for languages like C variants were focussed more on systems 
code (pointer chasing, etc.)  I think the math libraries were also better, 
complex number types were more of a first class citizen, etc.  In short, 
while the language was in many ways dated, it was being well maintained 
for a specific audience that other languages weren't addressing as well. 
In part for that reason, innovation continued on the FORTRAN platform 
certainly into the early '90s, and I suspect well beyond.  It's probably 
more like 15+ years ago, but folks like Ken Kennedy of Rice U. (who I'm 
very sad to say died just a week or two ago) were doing most of their 
parallel computing experimentation on a FORTAN base.  I suspect things 
have changed some since then.  More and more large compiler efforts share 
back ends and optimizers accross multiple source languages, suggesting 
that the optimizations for C languages and FORTRAN have probably gotten 
closer, etc.   No doubt a lot of parallel computing work, including for 
numeric codes, has moved to newer languages.  Still, there were reasons 
other than "lack of vision" why scientists have persisted on using FORTAN 
long after most CS types figure it's become petrified.  Maybe it has now, 
but if so that's relatively recent.

--------------------------------------
Noah Mendelsohn 
IBM Corporation
One Rogers Street
Cambridge, MA 02142
1-617-693-4036
--------------------------------------