@halirutan Any comments on the above before I actually start doing it?

@Szabolcs just to make doubly sure you saw it:

Take a look at the CArray docs:

It's quite disturbing to find such things ...

@Szabolcs Hm, is it so bad? Is it not a nice that every aspect of parsing a C expression can be represented using symbolic C?

Of course that doesn't mean everybody that uses symbolic C should use it

@JacobAkkerboom But it's not a two-dimensional array access ... and there doesn't seem to be a way to represent x,y on its own, without being surrounded by anything, even though it's valid C.

@JacobAkkerboom Looks very interesting, have you used it?

@Szabolcs Ah yes that is indeed disturbing

I have no used the package unfortunately.

Have you done more research on what type 1 might mean in LibraryLink (this answer you wrote)?

To anyone with pre V9 Mathematica: Plotting numeric functions (e.g. NIntegrate, x /. FindRoot[f[x],...]) does not need ?NumericQ protection from V9 onwards. Someone added a Plotting section to mathematica.stackexchange.com/questions/18393/…, some time back, and only now I've checked it. More precise version information about when it's needed might be helpful. Also, are there cases where it's necessary in V9+?

@Szabolcs Ah unfortunately also no. I'm glad you are also curious. But I haven't looked at any of this stuff in quite some time

I'm throwing in the towel, so: is anyone aware of a working implementation of fast multipole in Mathematica? I want to be really sure before I try writing an implementation this weekend…

None of my searches turned up practical code.

@Guesswhoitis. Just the result of some weakish googling, don't know if its useful to you: library.wolfram.com/infocenter/Articles/1438 . It does mention a package, but I don't see a link

@Jacob, thank you; I actually have a photocopy of that paper. Unfortunately the routines in the paper are skeletal, and still need a lot of modification for robustness… :(

Still, thank you!

Actually, my plan in the absence of better implementations was to suck it up and modify their routines… :D

Haha :)

I don't suppose there is real MMA code in the paper right?

There's code, but as I said, very bare, and will fail for nontrivial input.

Much as I like programming numerical routines, this is ridiculous… surely somebody had to do many-body simulations in Mathematica before, using FMM.

hmm

Ok, some more google fu: moniker.name/worldmaking/?p=81 and see "download: fmm mathematica files".

I've gotta go. See you

See you, @Jacob.

@blochwave I voted to close this question. Initially, it very much looked like a reoccurrence of the 10.0.0 bug (regressions occur you know). After Michael's post it seemed clear there was more to it and I voted to reopen.