Mike FABIAN wrote:> Matthias Clasen <mclasen@redhat.com> ????????: > > >>I secured evidence of broken caches being created occasionally >>by the fontconfig mmap snapshot in Fedora Rawhide. I won''t have >>time to look into the exact nature of the problem today, but >>I have put a tarball up here: >> >>http://people.redhat.com/mclasen/bad-caches.tar.gz >> >>It also contains the fonts that were used to create the caches >>on an i686 machine. The output of fc-cat looks as if the charset >>fields contain garbage.Actually, it''s not just the charset values: on the japanese/TrueType directory, all the strings, charset and langset values are broken. I''m rather confused as to how these font caches get generated though: I haven''t yet observed fc-cache generating a broken cache file. Has anyone seen fc-cache being bad, or is it some other program? pat
>>>>> "Patrick" == Patrick Lam <plam@MIT.EDU> writes:Patrick> Actually, it''s not just the charset values: on the Patrick> japanese/TrueType directory, all the strings, charset and Patrick> langset values are broken. Have you tried running fc-cache with different locales? (Just a guess....) BTW, I had to downgrade fc to before the patch that stores relative filenames in the cache-2 files. Without the full paths in the cache files nothing was able to actually open their fonts. Matching worked as intended, but the lib then tried to open those files in $CWD rather than in the same dir as the cache-2 file with the matching record. (Or so I think; in any case the verifiable result was a SEGV.) -JimC -- James H. Cloos, Jr. <cloos@jhcloos.com>
James Cloos wrote:>>>>>>"Patrick" == Patrick Lam <plam@MIT.EDU> writes: > > > Patrick> Actually, it''s not just the charset values: on the > Patrick> japanese/TrueType directory, all the strings, charset and > Patrick> langset values are broken. > > Have you tried running fc-cache with different locales? > > (Just a guess....) > > BTW, I had to downgrade fc to before the patch that stores relative > filenames in the cache-2 files. Without the full paths in the cache > files nothing was able to actually open their fonts. Matching worked > as intended, but the lib then tried to open those files in $CWD rather > than in the same dir as the cache-2 file with the matching record. (Or > so I think; in any case the verifiable result was a SEGV.)Yes, I had to revert that patch. We need to put full paths into cache files, but I''m not sure how to do that: does anyone know how to get the full path for a file? Actually, I wonder if that''s even correct if fonts get placed on different directories on different systems (do they?) I''ll try different locales. It''s weird. pat
Matthias Clasen wrote:> I secured evidence of broken caches being created occasionally > by the fontconfig mmap snapshot in Fedora Rawhide. I won''t have > time to look into the exact nature of the problem today, but > I have put a tarball up here: > > http://people.redhat.com/mclasen/bad-caches.tar.gz > > It also contains the fonts that were used to create the caches > on an i686 machine. The output of fc-cat looks as if the charset > fields contain garbage.Is it really different from the contents of fonts.cache-1, though? Upon visual inspection, I didn''t see that much of a difference. I''m continuing to look at this problem... pat
Matthias Clasen <mclasen@redhat.com> ????????:> I secured evidence of broken caches being created occasionally > by the fontconfig mmap snapshot in Fedora Rawhide. I won''t have > time to look into the exact nature of the problem today, but > I have put a tarball up here: > > http://people.redhat.com/mclasen/bad-caches.tar.gz > > It also contains the fonts that were used to create the caches > on an i686 machine. The output of fc-cat looks as if the charset > fields contain garbage.I think I see similar problems on SUSE Linux 10.1 Alpha. See also: http://bugzilla.novell.com/show_bug.cgi?id=128080 -- Mike FABIAN <mfabian@suse.de> http://www.suse.de/~mfabian ?????????????
I secured evidence of broken caches being created occasionally by the fontconfig mmap snapshot in Fedora Rawhide. I won''t have time to look into the exact nature of the problem today, but I have put a tarball up here: http://people.redhat.com/mclasen/bad-caches.tar.gz It also contains the fonts that were used to create the caches on an i686 machine. The output of fc-cat looks as if the charset fields contain garbage. Regards, Matthias