On macOS ARM, the build was failing because the file we wanted was in `/opt/homebrew/opt/freetype/include/freetype2/freetype/freetype.h` but the `#include` was looking for it in `/opt/homebrew/opt/freetype/include/freetype2/freetype2/freetype/freetype.h`

Nonetheless, it managed to work on x86_64 macOS, because on x86_64 macOS, Homebrew installs itself in `/usr/local`. There, the build was really using `/usr/local/include/freetype2/freetype/freetype.h`, with `/usr/local/include` implicitly on the include path.

On ARM macOS, Homebrew installs itself in `/opt/homebrew`, and makes `/opt/homebrew/include/freetype2/freetype/freetype.h` available via a symlink, but `/opt/homebrew/include` *isn't* implicitly on the include path.

I think (hope?) that the change I'm making here will work on Linux and other machines because `pkg-config` will point to the right place, but I've only tested this change on macOS 12 on x86_64 and ARM.

Since 2.53, frotz doesn't support the macOS built-in version of ncurses (5.4) unless you pass ITALIC= to disable italics. (ncurses 5.4 doesn't support italics.)

On macOS, we'll configure the build to rely on Homebrew's ncurses6-config.

Infocom stored a username in the same place that Inform stores its
version number.  Somehow it was discovered that the Z-machine would set
the username from the TOPS20 or Unix development environment, but it was
unknown at that time that any game made any use of it.  With the public
release of the Infocom source code, it was discovered that "Hollywood
Hijinx" used this functionality to enable a debug mode.  Because of this
clobbering when running V6 Z-code and that no other game seems to use
this, it's safe to completely ignore the username stuff.

Maybe it can be brought back in a roundabout way to put "Hollywood
Hijinx" into debug mode using the hotkeys.

This is for two reasons.  One is just for grins.  The other is for
investigating how to properly process EGA and CGA old-style graphics
files.

To get EGA properly framed, it seems that the thing to do is find
variables like gtable[i].width, gtable[i].height, data->width,
data->height, and divide by two.  Fiddling around with the eight places
I found in x_oldpic.c to do this, I managed to get width okay, but not
height.  I'm having a hard time identifying what to alter and what to
leave alone or if I need to look at other variables too.

I'm unsure if this should be in Frotz, so I'm removing it until I can
fully investigate why this is desirable and how abuses can be prevented.

This reverts commit 2f621630.

This is part of the Z Machine Specificiations version 1.1.
See https://inform-fiction.org/zmachine/standards/z1point1/sect15.html#save

This entailed switching from the Frotz-contained zgetopt() to the
OS-provided getopt_long().  There were no sensible single-letter flags
available for this.  Given that I don't anticipate SDL Frotz being built
on any machine that doesn't have getopt_long() and/or getopt.h
available, this is a reasonable choice.  If someone wants to submit a
patch to implement long options, I'm open to adding it

Lurking Horror with sound was released only for the Amiga.  That version
assumes that the interpreter is very slow, so it fires off several sound
effects in one round assuming there will be time to play most of each
one.  What happens on a fast modern machine is the sound starts and is
immediately stopped, so they don't appear to play at all.

See the "Remarks" section in
http://inform-fiction.org/zmachine/standards/z1point1/sect09.html

When the SDL Frotz code was originally written or converted to SDL 2,
this caveat was forgotten.