[Groff] Composite output regression due to Unicode fonts?

[Colin Watson]

I've been testing out how 1.19.2 and CVS groff with input files in
various languages and encodings fed through preconv. Using the attached
troff output file generated by:

  zcat /usr/share/man/pl/man1/822-date.1.gz | ./src/preproc/preconv/preconv 
-eISO-8859-2 | ./test-groff -mandoc -Tutf8 -Z

... on a Debian system with manpages-pl installed and with CVS groff, I
found that 1.19.2's grotty correctly renders characters such as U+017C
and U+0105 (which appear as u007A_0307 and u0061_0328 respectively in
the troff output), but CVS's grotty displays these as only the base
character (z and a respectively). Characters not rendered as Unicode
composite characters, such as U+0107 => 'c, display correctly using both
versions of groff. The attached diff should demonstrate the problem
clearly.

Furthermore, running CVS's grotty with 1.19.2's fonts restores the
previous handling.

It seems as if this problem might be due to Unicode fonts. Is it
possible that libdriver doesn't know how to deal with composite
characters that aren't explicitly listed in the font definition?

Thanks,

-- 
Colin Watson                                       address@hidden

822-date.pl.ditroff
Description: Text document

822-date.pl.out.diff
Description: Text document