Bug in font subsetting

[JSWolf]

Seems the font subsetting is removing ligatures and ADE for Windows is trying to render them from non-ligature text. So what we have are missing charters since the ligatures are now gone.

Is this something that's easy enough to fix?

[kovidgoyal]

You need to provide an example book, fonts are complex, there could be any number of things going on.

[JSWolf]

Originally Posted by kovidgoyal

You need to provide an example book, fonts are complex, there could be any number of things going on.

The current version of ADE converts characters such as fl ff fi etc. into ligatures. When the embedded font is subset, the ligatures are removed and the rendering is thus not correct.

The ePub I've attached shows a subsetted font and missing letters due to ligatures.

When Rand and Mat carried the first

shows as

When Rand and Mat carried the  rst

because the fi in the word first is converted to a ligature and the ligature is not in the subsetted font. Calibre's viewer also show's the same bug.

[kovidgoyal]

calibre's font subsetting code actually reads the gsub table in a font and includes all ligature combinations of the subset glyphs. The test file you attached has no f based words and no embedded fonts.

And converting a text file with the sentence When Rand and Mat carried the first in it with the embedded font Calibri (which has a fi ligature) results in the same rendering with and without subsetting in the calibre viewer.

EDIT: Nevermind, I see that I downloaded an incorrect file, I'll look at your file in a bit

[JSWolf]

The font in use is Charis SIL. The link below is to the full versions of the font.

http://www.mediafire.com/download.php?64cse2dshispx08

[JSWolf]

Originally Posted by kovidgoyal

calibre's font subsetting code actually reads the gsub table in a font and includes all ligature combinations of the subset glyphs. The test file you attached has no f based words and no embedded fonts.

And converting a text file with the sentence When Rand and Mat carried the first in it with the embedded font Calibri (which has a fi ligature) results in the same rendering with and without subsetting in the calibre viewer.

EDIT: Nevermind, I see that I downloaded an incorrect file, I'll look at your file in a bit

Please let me know when a fix is in place. Thank you.

[kovidgoyal]

Your problem is more likely caused by the font-family declarations using the same font-family name for all four variants of the font. Change it to use different names and you should be fine. I suspect there is something about the way the charis compact font is designed that causes this usage to be problematic. If not, let me know.

[JSWolf]

Originally Posted by kovidgoyal

Your problem is more likely caused by the font-family declarations using the same font-family name for all four variants of the font. Change it to use different names and you should be fine. I suspect there is something about the way the charis compact font is designed that causes this usage to be problematic. If not, let me know.

The problem is not with the font or the font declaration. The problem is with Calibre. The problem is that Calibre doesn't see any ligatures so the ligatures get removed. Then ebook-viewer and ADE 2.0 both take the regular characters such as fi and fl and try to output ligatures. There are no ligatures in the subsetted font since Calibre has wiped them out. So, what we get is space instead of the ligature(s).

The way to do it (IMHO) is to make a list of all the ligatures and if the pairs exist in the ePub (unligatured), keep the ligature. As it is now, the only way the ligatured is kept is if the ligature is in the text.

[kovidgoyal]

You're right, there was a bug in the parsing of the GSUB table that was triggered by this particular font. Fix will be in the next release.

[JSWolf]

Originally Posted by kovidgoyal

You're right, there was a bug in the parsing of the GSUB table that was triggered by this particular font. Fix will be in the next release.

Thank you very much!