actually the output array doesn't need to be there at all, I
could compress on the fly, adding the filter byte to a much
smaller buffer, then writing straight out to the file. Then the
only big array would be the one in the image class itself, which
is easy to manage.
In fact, I did another file originally called imagedraft.d that
does pngs as lazy ranges. Doesn't look like I ever posted that
one to github. I think it can actually read more png files too.
But in any case, switching to that will be harder since I don't
even remember how to use it.
ANYWAY, I added a new function to png.d:
void writeImageToPngFile(in char[] filename, TrueColorImage
image);
Use it like this:
// and finally write the data to a png file
writeImageToPngFile("images/"~toHexString(md5Of(curtrans))~".png",
i);
And see if things are any better. This one uses malloc/free for
its temporary buffer and uses the Compress class from std.zlib to
do it one line at a time instead of the big function, so
hopefully its temporary buffer will be more manageable too.
It also explicitly deletes the compressed data buffer when it is
done with it, after writing to the file. (I know delete is
deprecated but switching this to use an malloced buffer would
have needed me to totally duplicate the std.zlib class so meh, we
can do it if needed but for now let's just try this).
Since my computer can't handle the large arrays anyway, I can't
test this for sure but I expect you'll see some difference here.
If you've changed TrueColorImage to use malloc/free as well that
would probably help as well (none of this keeps a reference to
data, so free it in truecolorimage's destructor and you should be
ok.)
