Let's try using fdopen() and feed it a FILE * instead.
This also avoids trouble with the g++ 3.x -> 4.x interface change for
the 'int fd' constructor.

Make kira_open read plain files again.

Don't discard texture data after loading it into opengl.
If the window gets recreated (detached, quadbuffer-mode-switched, etc.),
we'll need it again.

XXX Or, should we just re-read the image files if the occasion arises?

or when it changes quadbuffered-stereo status.

which it uses.  Doesn't work yet -- not sure why -- appears the
streambuf isn't set up properly (it always returns EOF).

Use GNU-C++ extension <ext/stdio_filebuf.h>, which can turn
a unix file descriptor into a streambuf.

that they don't work.  Use an istream rather than ifstream -- we should
be able to create an istream from a socket, and that's all the starlab libs need.

Work in a way compatible with locking.

to make partibrains.c a hair less bloated.
Add new locking functions, specks_lock()/specks_unlock().
Don't pass specklist **'s among non-locking functions, so e.g.
new specks_ensuretime() replaces specks_timespecksptr().

don't env-expand if first nonblank char is "#".

AC_CANONICAL_HOST doesn't set anything mentioning MINGW32; instead
it seems to set OSTYPE to "msys".  So accept either that or mingw32.

Add THREAD_* flags.
Rearrange CFLAGS, etc., so that it's easier to do things like
make OPT="-g" to select debugging without optimization, etc.

unfortunately linux-specific (opens /dev/fd/*).

Oh, and the new option is called --enable-threads (not --enable-pthreads).

Add possibly-misguided ThReader structure.

Try different way of detecting opengl libs in mingw32 case too.

bug, so remove the Apple-specific workaround.

Add "ready" parameter to dyndata trange method.

which seems to have bugs on some implementations (?!).
Let's see if this helps.

Add env var PARTIOLDOPENGL to force using non-glDrawElements path.