If DigitalMan uses the Edit and Continue feature, he can fix it.  :-)

sdl_video_event_thread().

This should fix the crash on startup bug seen on Windows systems.

When built withOUT the preprocessor macro WITH_SDL defined, ciolib.h doesn't
include gen_defs.h (perhaps it should), so the use of FREE_AND_NULL in this
file become function calls and the symbol is not found when linking.
This is why jsexec was not included in recent daily dev builds of v3.16.

LFs on write.

the mod state.  This should fix num-lock on Windows.

Other mod states (CTRL, SHIFT, ALT) are likely still messed up though.

any line that was ever damaged was always damaged, and it worked pretty well.

when updating the screen.

This should fix the weird long-delayed update issue.

the screen.  This fixes the phantom cursor issue.

For regular vstatlock things, only hold the lock long enough to copy the
variables that will be used.

For vstat.vmem stuff, create a separate reference-counting vmem structure
and a new vmem_lock so the memcpy() of the vmem won't block basic vstat
accesses.

This appears to fix the recently-added "blockiness" of screen updates in
SyncTERM, without risking deadlocks or unlocked accesses... the major issue
holding up a release.

issues.

and release it last.

hold the lock.  This should finally fix all the things I hope.

already held by the bitmap layer.

This fixes the Linux crash on exit in SDL when X11 mode is used (and possible
others).

a potential problem with grabbing vstatlock).  Call XCloseDisplay and terminate
the X11 event thread on exit().

This still does not prevent the crash on exit for Linux.  This appears to be
an issue with SDL.

setup_surfaces() was never called for overlay mode, resulting in the window
contents never changing size.

scaling but forces it to 2.  This is to allow two "pixels" for each "real pixel".

If we don't have two pixels, the chrominance values cannot be represented
correctly, only the luminance... resulting in a monochrome image.

Valgrind report).

for the four current fonts... -99 (uninitialized), and -1 (loaded font).

-99 was supposed to be automagically converted to CP437, but a bug in
revision 1.40 (Feb 27, 2015) made this not happen.  As a result, the font
was copied from an index of -99 in the font data array.  This was garbage
on some platforms and appears to have caused crashes in others.

really may to what is listed.

video stats.  This makes the C64 80 column mode look less stupid.

new_scaling and old_scaling to prevent setup_surfaces() from being called
again while the window is initially being set up in a mode other than the
default 80x25.

Fixes a Windows crash when starting in other modes.

single argument.  Imapged Global War continent maps at least.

Thanks for the bug report Digital Man!

couple global pointers (win and new_rect), and move upd_rect, rectspace, and
rectsused into automatic variables in the event thread function.

the high intensity attribute can.  Using CTerm, this is controlled via the
CSI [ ? 34 h/l commands.

Also add CSI [ ? 35 h which disables blinking (reversed by CSI [ ? 35 l)

When a resize event is serviced, cache the new scaling factor.  Next time
the event queue is empty, attempt to lock vstatlock.  If that succeeds,
apply the new scaling.  If not, wait for the queue to drain again.

This may cause the scaling to be slightly delayed, but should completely
solve this deadlock.

This deadlock likely never affected FreeBSD since FreeBSD has deadlock
detection, and the return value of pthread_mutex_lock() isn't tested.  The
result would have been a silent failure to lock, and an update of the scaling
factor regardless, followed by a silent (again untested) failure of the matching
pthread_mutex_unlock() call.

by sdl_ufunc_mtx.  Fix up the sdl_ufunc_mtx usage to help prevent errors.

Also, work around a deadlock on Win32 where a video resize event while a
redraw event is later in the queue would deadlock on the vstatlock mutex.

This problem is *not* completely resolved yet.  I think the new scaling
will need to be cached, then applied the next time the queue is empty.

(the default).  If the program doesn't already handle CIO_KEY_EXIT, waiting
a half-second isn't going to help, and if it does, it should turn off the
reaper.