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.