Just on plugins and Mata and such, Mata also has one major advantage
over C. Compiling via Stata means that the source code can be
practically distributed rather that the binary code, which means not
only cross-platform stability, but security through transparency (not
that everybody will read the code, but just because they can).
Actually there are some additional advantages of Mata. One can
distribute an .mo file (or a .mlib library file) which although
binary is interpretable on all platforms, like a binary .dta file. As
Bill Gould said on this list last week, if you had > 2000 lines of
Mata code, you probably would not want it in-line in an ado-file. If
you have a routine (such as a 3-D graphics package--anyone interested
in porting one of those?!?) that is composed of many, many Mata
functions, you probably want to distribute an .mlib. That does not
prevent you from transparently distributing the source code as well,
but it would not need to be compiled on the fly as in-line code would.
Another advantage---which goes rather against the open-source spirit
of Statalist and SSC---is that Mata (like C) makes it *possible* to
distribute binary-only code for a routine. That makes it feasible for
someone to sell that code. (Indeed, that is the case for StataCorp-
authored commands like xtmixed). Personally I hope that practice will
not be widespread among the user community, but if someone had put
immense effort into developing something, they might not want to give
it away (e.g., James Davidson's TSMOD package, which recently went
from free to a modest charge; the flip side, of course, is that if
you pay for it you are paying for support).
