The unexpected effectiveness of Python in science

In the 1980's we had code that was from the 1950's and 1960's that we were converting from Fortran IV to Fortran77. [Most of those conversions was 'fix it so it compiles and gives back the same data on the test run'. So some of that Fortran you see is actually much much older.

A lot of this old-code has to do with amount of time non-coders have to deal with code, but part of it is that you may need to run years old data plus new data and need it to have the same answers as before. Or that the satellite you are using was launched in the 80's and is still producing data 40 years later. You could get some people in to recode stuff with a 40% chance of it all falling apart from budget cuts.. or you could use the old hardware and use the money to buy some lab equipment that got cut from a different budget.

This is going to be one of the reasons that Python2.7 is going to have a longer shelf life than people want or expect. A lot of scientists love 2.7 because it is not going to change. Every new RHL/Debian/etc release, we would have to go and deal with support tickets at the Lab due to changes in python not always being as backwards compatible as thought. [My understanding is that there are systems at CERN and similar labs running RHL-7.3 15 years later because it has the python that the experiments were written against.. it will still be running them for N more years until the experiment completes.] Most of the scientists running long lived experiments are writing the stuff in 2.7 because of the fact for them it is no longer moving. If there is ever a 3.N that stops they will move to that also (while still running 2.7 and 1.5 systems.)

At this point I expect an Numerical Recipes book in Python would be written against Python 2.7