[Haskell-cafe] Haskell for Physicists

[Khudyakov Alexey]

В сообщении от Среда 30 сентября 2009 23:08:02 вы написали:
> Yep, sure did.  I just hit `reply' assuming haskell-cafe was in the
> reply-to.  I do that more often than not it seems.
> Going back to the OP, what area of physics, and how on earth are you
> going to convert years of fortran users to haskell?
> I mean, in particle physics (were I came from) it seems as though only
> recently have they moved from fortran to C++ (note: C was skipped).
> There are things written in python (like Athena) but, well..., they are
> unreliable crap (I do like python though).
> In fact, when I was in undergraduate, not 4 years ago, a PhD student was
> writing his big QCD project in fortran from the ground up.  I'm not even
> familiar enough with fortran to attempt such a thing (I would have used
> C).  Case in point, I think there are some areas of physics that exist
> as a communal project (i.e. experimental particle physics) and because
> of this, you are limited to the tools and data used by your peers
> (Athena, Geant4, etc...).  It is really hard to introduce anything new.
> So I guess my advice would be to avoid Haskell as a 'replacement' for
> anything to a physicist (including mathematica -- which I never liked
> myself).  They will immediately ignore you.  Approach it as a new tool,
> and focus on what it can do that software-x can't.
> 
I'm particle physicist too. And sometimes I think that it would be better if 
they stay with fortran. Object-disoriented which is done in C++ scares me. 
Random segfaults in ROOT, or even worse segfault loops...

It's possible to use safety as argument for haskell. Type safety, no 
segfaults.

As for existing code there are two strategies. 

First is to dump all code into Geneva lake. There are environmental concerns 
of course. And it's difficult to throw away "tested" code.

Second one - do not touch it and use haskell for small isolated tasks. It's 
easier to do this in smaller experiments. I use haskell to process 
experimental data with reasonable success. Code is much cleaner and easier to 
understand that C++ code.

In fact I just reworded your statement