Subj : Physicists as programmers
To   : comp.programming
From : Jyrki Alakuijala
Date : Wed Jul 27 2005 11:47 am

Walter wrote:

> Many of the best programmers I know have never taken a course in
> programming. They have degrees in physics.

Many of the _worst_ programmers I know have a physics graduate degree.
They are inventive enough to return the address of a stack variable
or indecisively abstract the array indexing to avoid using plain
0 or 1 (because FORTRAN uses 1 and C uses 0 and they cannot stand
the conflict). If one is bright enough to code without knowing the
culture and tradition of coding, the resulting code almost works and
is close to unmaintainable.

Also, I have seen safety critical code produced by a physicist that
fraudulently produced its output. It was supposed to calculate a
verification image based on its computation, originated from an
input image. Unable to know how to calculate the image, but not
having the humility to admit his incapabilities in programming,
the physicist had decided to take the input image, smooth it
slightly, and give the smoothed input image as the verification
output. The level of "pragmatism" was overly high here, and I decided
to rewrite all the code he had delivered (during 1.5 years), taking a
whole two weeks for me. The resulting code was faster, allowing me
to run 100 iterations (instead of the earlier 3), only a quarter in
kLocs, had a more complete feature set, was more accurate and, most
importantly, did not forge its verification output.

Physicists, not knowing the testing culture in SE, typically do
a lot of manual testing of their code. Of course, a lot of
"improvements" are introduced between the successful tests, often 
breaking the previously tested features. Once the physicist's
code is transferred to SE people, they need to start by writing
the requirements, creating the automated test suite, only to find
that it does not work at all. The physicist people have covered
their back by producing binders full of reports about their
testing results, run manually during the development phase,
showing a complete agreement with the calculation results and the
measurements.

I would not allow the typical physics graduate to do any commercial
programming without a real programmer overseeing what is getting
done.

One of the best programmers I know has a Ph.D. in physics. However,
more than 95 % are really dangerous or slowish programmers. My
sampling is based mostly on the European professional programmer
population. ;-)

.