| Subject: | Certain fmt-4 cmaps rejected by Windows |
Some TrueType libraries (including that used by Windows font installer and previewer) requre that format-4 cmap subtables end with a sentinel segment that both starts and ends with 0xFFFF (not just ends with 0xFFFF as the current spec says).
Since 1.03, when writing cmap tables, Font::TTF::Cmap optimizes format-4 subtables and may combine 0xFFFF with preceding characters into the last segment. In this case the Windows installer and previewer will complain that the font is not a valid font.
Attached patch will force Cmap.pm to put 0xFFFF into its own segment. This has already been committed to the source repo.
| Subject: | cmap4_patch.patch |
Index: Font-TTF/trunk/lib/Font/TTF/Cmap.pm
===================================================================
--- Font-TTF/trunk/lib/Font/TTF/Cmap.pm (revision 999)
+++ Font-TTF/trunk/lib/Font/TTF/Cmap.pm (revision 1000)
@@ -416,7 +416,14 @@
{
my (@starts, @ends, @deltas, @range);
- push(@keys, 0xFFFF) unless ($keys[-1] == 0xFFFF);
+ # There appears to be a bug in Windows that requires the final 0xFFFF (sentry)
+ # to be in a segment by itself -- otherwise Windows 7 and 8 (at least) won't install
+ # or preview the font, complaining that it doesn't appear to be a valid font.
+ # Therefore we can't just add 0XFFFF to the USV list as we used to do:
+ # push(@keys, 0xFFFF) unless ($keys[-1] == 0xFFFF);
+ # Instead, for now *remove* 0xFFFF from the USV list, and add a segement
+ # for it after all the other segments are computed.
+ pop @keys if $keys[-1] == 0xFFFF;
# Step 1: divide into maximal length idDelta runs
@@ -473,6 +480,12 @@
# Finished with this macro-range
$start = $end;
}
+
+ # Ok, add the final segment containing the sentry value
+ push(@keys, 0xFFFF);
+ push @starts, 0xFFFF;
+ push @ends, 0xFFFF;
+ push @range, 0;
# What is left is a collection of segments that will represent the cmap in mimimum-sized format 4 subtable