Enhance icon.pl to be a fully general Windows icon builder: it now

supports monochrome icons, can deal with any size of image you're
mad enough to put in there, and will construct icons with whatever
combination of sizes and resolutions you feel like specifying. This
has involved a change in the command-line syntax, hence the
adjustment to Makefile.

(I don't imagine that the changes described here will be critical to
Puzzles any time soon, but I might reuse this script elsewhere and
then I won't want it to have arbitrary limitations.)

[originally from svn r7031]

icons/Makefile

@@ -117,7 +117,9 @@ $(P16D4): %-16d4.png: %-ibase.png
 $(ICONS): %.ico: %-48d24.png %-48d8.png %-48d4.png \
                  %-32d24.png %-32d8.png %-32d4.png \
                  %-16d24.png %-16d8.png %-16d4.png
-	$(PIC)icon.pl $? > $@
+	$(PIC)icon.pl -24 $*-48d24.png $*-32d24.png $*-16d24.png \
+		      -8  $*-48d8.png  $*-32d8.png  $*-16d8.png  \
+		      -4  $*-48d4.png  $*-32d4.png  $*-16d4.png  > $@
 
 # Build the .RC files which bind the icons into the applications.
 $(RC): %.rc:

icons/icon.pl

@@ -1,23 +1,58 @@
 #!/usr/bin/perl 
 
-# Take nine input image files and convert them into a
+# Take a collection of input image files and convert them into a
-# multi-resolution Windows .ICO icon file. The nine files should
+# multi-resolution Windows .ICO icon file.
-# be, in order:
 #
-#  - 48x48 icons at 24-bit, 8-bit and 4-bit colour depth respectively
+# The input images can be treated as having four different colour
-#  - 32x32 icons at 24-bit, 8-bit and 4-bit colour depth respectively
+# depths:
-#  - 16x16 icons at 24-bit, 8-bit and 4-bit colour depth respectively
 #
-# ICO files support a 1-bit alpha channel on all these image types.
+#  - 24-bit true colour
+#  - 8-bit with custom palette
+#  - 4-bit using the Windows 16-colour palette (see comment below
+#    for details)
+#  - 1-bit using black and white only.
 #
-# TODO: it would be nice if we could extend this icon builder to
+# The images can be supplied in any input format acceptable to
-# support monochrome icons and a user-specified subset of the
+# ImageMagick, but their actual colour usage must already be
-# available formats. None of that should be too hard: the
+# appropriate for the specified mode; this script will not do any
-# monochrome raster data has the same format as the alpha channel,
+# substantive conversion. So if an image intended to be used in 4-
-# monochrome images have a 2-colour palette containing 000000 and
+# or 1-bit mode contains any colour not in the appropriate fixed
-# FFFFFF respectively, and really the biggest problem is designing
+# palette, that's a fatal error; if an image to be used in 8-bit
-# a sensible command-line syntax!
+# mode contains more than 256 distinct colours, that's also a fatal
+# error.
+#
+# Command-line syntax is:
+#
+#   icon.pl -depth imagefile [imagefile...] [-depth imagefile [imagefile...]]
+#
+# where `-depth' is one of `-24', `-8', `-4' or `-1', and tells the
+# script how to treat all the image files given after that option
+# until the next depth option. For example, you might execute
+#
+#   icon.pl -24 48x48x24.png 32x32x24.png -8 32x32x8.png -1 monochrome.png
+#
+# to build an icon file containing two differently sized 24-bit
+# images, one 8-bit image and one black and white image.
+#
+# Windows .ICO files support a 1-bit alpha channel on all these
+# image types. That is, any pixel can be either opaque or fully
+# transparent, but not partially transparent. The alpha channel is
+# separate from the main image data, meaning that `transparent' is
+# not required to take up a palette entry. (So an 8-bit image can
+# have 256 distinct _opaque_ colours, plus transparent pixels as
+# well.) If the input images have alpha channels, they will be used
+# to determine which pixels of the icon are transparent, by simple
+# quantisation half way up (e.g. in a PNG image with an 8-bit alpha
+# channel, alpha values of 00-7F will be mapped to transparent
+# pixels, and 80-FF will become opaque).
 
+# The Windows 16-colour palette consists of:
+#  - the eight corners of the colour cube (000000, 0000FF, 00FF00,
+#    00FFFF, FF0000, FF00FF, FFFF00, FFFFFF)
+#  - dim versions of the seven non-black corners, at 128/255 of the
+#    brightness (000080, 008000, 008080, 800000, 800080, 808000,
+#    808080)
+#  - light grey at 192/255 of full brightness (C0C0C0).
 %win16pal = (
     "\x00\x00\x00\x00" => 0,
     "\x00\x00\x80\x00" => 1,
@@ -38,18 +73,32 @@
 );
 @win16pal = sort { $win16pal{$a} <=> $win16pal{$b} } keys %win16pal;
 
+# The black and white palette consists of black (000000) and white
+# (FFFFFF), obviously.
+%win2pal = (
+    "\x00\x00\x00\x00" => 0,
+    "\xFF\xFF\xFF\x00" => 1,
+);
+@win2pal = sort { $win16pal{$a} <=> $win2pal{$b} } keys %win2pal;
+
 @hdr = ();
 @dat = ();
 
-&readicon($ARGV[0], 48, 48, 24);
+$depth = undef;
-&readicon($ARGV[1], 48, 48, 8);
+foreach $_ (@ARGV) {
-&readicon($ARGV[2], 48, 48, 4);
+    if (/^-(24|8|4|1)$/) {
-&readicon($ARGV[3], 32, 32, 24);
+	$depth = $1;
-&readicon($ARGV[4], 32, 32, 8);
+    } elsif (defined $depth) {
-&readicon($ARGV[5], 32, 32, 4);
+	&readicon($_, $depth);
-&readicon($ARGV[6], 16, 16, 24);
+    } else {
-&readicon($ARGV[7], 16, 16, 8);
+	$usage = 1;
-&readicon($ARGV[8], 16, 16, 4);
+    }
+}
+if ($usage || length @hdr == 0) {
+    print "usage: icon.pl ( -24 | -8 | -4 | -1 ) image [image...]\n";
+    print "             [ ( -24 | -8 | -4 | -1 ) image [image...] ...]\n";
+    exit 0;
+}
 
 # Now write out the output icon file.
 print pack "vvv", 0, 1, scalar @hdr; # file-level header
@@ -65,13 +114,15 @@ for ($i = 0; $i < scalar @hdr; $i++) {
 
 sub readicon {
     my $filename = shift @_;
-    my $w = shift @_;
-    my $h = shift @_;
     my $depth = shift @_;
     my $pix;
     my $i;
     my %pal;
 
+    # Determine the icon's width and height.
+    my $w = `identify -format %w $filename`;
+    my $h = `identify -format %h $filename`;
+
     # Read the file in as RGBA data. We flip vertically at this
     # point, to avoid having to do it ourselves (.BMP and hence
     # .ICO are bottom-up).
@@ -135,33 +186,40 @@ sub readicon {
 	die "too many colours in 8-bit image $filename\n" unless $palindex <= 256;
     } elsif ($depth == 4) {
 	%pal = %win16pal;
+    } elsif ($depth == 1) {
+	%pal = %win2pal;
     }
 
     my $raster = "";
     if ($depth < 24) {
 	# For a non-24-bit image, flatten the image into one palette
 	# index per pixel.
-	my $currbyte = 0, $currbits = 0;
+	$pad = 32 / $depth; # number of pixels to pad scanline to 4-byte align
-	for ($i = 0; $i < scalar @$data; $i++) {
+	$pmask = $pad-1;
-	    $pix = $data->[$i];
+	for ($y = 0; $y < $h; $y++) {
-	    $currbyte <<= $depth;
+	    my $currbyte = 0, $currbits = 0;
-	    $currbits += $depth;
+	    for ($x = 0; $x < (($w+$pmask)|$pmask)-$pmask; $x++) {
-	    if (defined $pix) {
+		$currbyte <<= $depth;
-		if (!defined $pal{$pix}) {
+		$currbits += $depth;
-		    die "illegal colour value $pix at pixel $i in $filename\n";
+		if ($x < $w && defined ($pix = $data->[$y*$w+$x])) {
+		    if (!defined $pal{$pix}) {
+			die "illegal colour value $pix at pixel $i in $filename\n";
+		    }
+		    $currbyte |= $pal{$pix};
+		}
+		if ($currbits >= 8) {
+		    $raster .= pack "C", $currbyte;
+		    $currbits -= 8;
 		}
-		$currbyte |= $pal{$pix};
-	    } else {
-		$currbyte |= 0;
-	    }
-	    if ($currbits >= 8) {
-		$raster .= pack "C", $currbyte;
-		$currbits -= 8;
 	    }
 	}
     } else {
 	# For a 24-bit image, reverse the order of the R,G,B values
 	# and stick a padding zero on the end.
+	#
+	# (In this loop we don't need to bother padding the
+	# scanline out to a multiple of four bytes, because every
+	# pixel takes four whole bytes anyway.)
 	for ($i = 0; $i < scalar @$data; $i++) {
 	    if (defined $data->[$i]) {
 		$raster .= $data->[$i];