This.

Basically, Windows for scientific software is as bad as Linux was for 3D FPS gaming fifteen years ago. Just ask anyone who has tried compiling anything scientific under Cygwin.

And don't get me started on MS's own compilers. They STILL don't support C99, a sixteen year old standard for one of the most important programming languages!

I thought most of the scientific computation is written in Fortran. Anyways, I have used a big closed source scientific computation package several jobs ago to conduct experiments. The entire stack was running on windows at the time, but the operating system really did not (and does not) matter. I agree with you that MS compilers should support C99 but it has very little to do with how useful windows is for scientific computing. (Worst case scenario you can use an opensource compiler).

I think the reason why the top 500 HPC clusters are using linux is the price of the operating system and not the functionality.

What do you mean by scientific? There's a lot of Windows only software for research. Mostly for applied things. But also things like data collection. There's even more if you look at development in things like engineering. Even industrial is surisingly much Windows only.

Well, yes, I'll give you that "scientific software" is indeed too broad. If I restrict it to "software developed by teams of scientists working at universities and research labs" it's better.

And yes, lots of data aquisition stuff is Windows only, particularly the stuff that gets sold along instruments. But then, in that category, the trend seems to be requiring a (preferrably air-gapped) PC running XP. I've even seen stuff running only on Win98 still in use, and I've seen software for rheometers that is STILL distributed on 3.5" floppies. And don't get me started on the "needs-crypto-usb-dongle-to-function" stuff - someone should really tell them that the $300k instrument is a sufficient protection against software piracy.

Makes sense. My concern really is if Linux gets any better for these use cases. I think a lot of Linux development is to much software and to little real world "science". You hear a lot how powerful Linux is, but in a lot of cases powerful things are run by XP as you say.

I kind of gave on Linux a couple of years ago when at the same time when Apple had totally neglected the Mac Pro all you could find from the Linux community was how cool the latest 3D effects on the desktop where and chromium was much better than Firefox. Trying instead finding something like say a 2D library more modern than Cairo and there's a deafening silence.

Yes, I'm exaggerating a bit and there is some good developments. Like CERN making waves in electronics by hacking on KiCad. Though I'm sure they would be happier if they could replace Wx with something more modern, which I'm not sure the current alternatives are.

From what I gather, Qt has "won" the war, if there was ever such a thing, but GTK and Cairo will not be going away. What are your gripes with Cairo, anyways?

Also, when you start getting into scientific software, very few projects explicitly state "Runs best on Linux" even though that may be the case. If you are looking for examples though, I will give some: PETSc/Elemental/Trilinos/etc Clawpack/PyClaw/OpenFOAM/etc Gromacs/NAMD/HOOMD/Quantum Espresso/etc Julia/Nimrod/etc Paraview/VisIt/VMD/etc OpenMPI/MPICH/MVAPICH/etc ImageJ/Fiji/etc GNU Radio/RTL-SDR/etc

There's tons of it out there!

You seem to use Linux for applications. It appears you have nothing to do with research about the Linux source code and design.

So? The Linux ecosystem is the best there is for science. Period. Even OSX lags behind, just look at all the work people put into MacPorts and Homebrew just to get it almost on par with Linux. To this day, none of those systems properly handle the case of multiple packages providing implementations of a library that other packages depend on (e.g. OpenMPI vs. MPICH). Not to talk down on the guys working on that, they are very talented, and the amount of effort they put in is of huge value to the community. (Hello Sean, hello Geoff.)