🚀 go-pugleaf

In article <C1BEIn.8MA@umassd.edu> hshubs@cis.umassd.edu (Howard S Shubs) writes:
>In <1993Jan23.151052.6593@hubcap.clemson.edu> jtbell@hubcap.clemson.edu (Jon Bell) writes:
>
>>It's already been done.  When I was a grad student in experimental
>>particle physics, I worked with CERN's HYDRA library, which implemented
>>dynamic memory allocation in FORTRAN using the same basic idea.  I 
>>think this package is still widely used in particle physics.
>
>That's meaningless.  Just because something has been done poorly is not
>a reason to continue to do it poorly.  Someone has to maintain that code.

The CERN HYDRA package first appeared in the late 60's, I think.  Its goal
was to provide linked data structures and dynamic memory management in
the context of standard Fortran (at that time, Fortran 66) so that it could
be run on as many platforms as possible.  Particle physicists work in large
collaborations, on a wide variety of computers, and source-code
portability is important to them.  

In in ideal world, I suppose the best solution would have been to migrate
to another language which had built-in support for dynamic memory allocation.
But at the time, I don't think there _was_ any other suitable language.
Algol was basically used only in Europe.  PL/I was used mostly on IBM
machines.  C and Pascal hadn't even been invented yet!

Now, of course, we _do_ have widely-used languages which have dynamic 
memory allocation, but physicists still love Fortran, and there's an awful
lot of code out there.

Actually, HYDRA was pretty well documented.  Most of the problems in using
it were the standard conceptual ones involving linked data structures and
dynamic memory management, dangling pointers and the like.  And since it
was a very widely used package, you could always find somebody who 
understood it and its coding conventions.  Maintenance wasn't _that_ big
a problem.

I certainly wouldn't recommend this approach nowadays for a new software
system.  If it needed dynamic memory allocation, I'd say, "Use C... or
Pascal... or maybe Fortran 90.  Don't even think about doing it in 
Fortran 77."  But for its time, HYDRA was the only possible solution.

Jon Bell / Dept. of Physics & C.S. / Presbyterian College / Clinton SC