Red Hat Bugzilla – Bug 491596
libmagic does not recognize pcf.gz files as font files
Last modified: 2009-03-23 11:44:11 EDT
libmagic does not recognize pcf.gzfiles as font files
This is necessary for font autoprovides to work with pcf.gz files
+++ This bug was initially created as a clone of Bug #491510 +++
The new font autoprovides are working well for some font formats but are not processing others:
1. ttc files: http://koji.fedoraproject.org/koji/rpminfo?rpmID=1026593
2. pfa files: http://koji.fedoraproject.org/koji/rpminfo?rpmID=1126190
3. compressed pcf fonts: http://koji.fedoraproject.org/koji/rpminfo?rpmID=1054305
(and probably plain pcf fonts though I could not locate a pcf package that was properly converted yet)
--- Additional comment from email@example.com on 2009-03-22 08:52:40 EDT ---
Also, it seems to fail even for plain ttf files, when they have a space in their name: http://koji.fedoraproject.org/koji/rpminfo?rpmID=1055798
--- Additional comment from firstname.lastname@example.org on 2009-03-23 02:28:14 EDT ---
The problem with .ttc and .pfa is that neither is seen as font data by libmagic:
[pmatilai@localhost cjkuni-ukai-fonts-0.2.20080216.1]$ file ukai.ttc
[pmatilai@localhost culmus-0.102]$ file YehudaCLM-Bold.pfa
YehudaCLM-Bold.pfa: PostScript document text
These should be treated filed against "file" (libmagic), rpm will recognize them as fonts as soon as libmagic does.
Compressed files aren't currently being looked into. I initially enabled it when enabling the font provide stuff but it blew up at least openoffice.org build due to leaking file descriptors someplace (probably libmagic but haven't had a chance to look at it).
Spaces in filenames causing issues is likely to be something entirely different, lets not pile things up.
--- Additional comment from email@example.com on 2009-03-23 04:31:25 EDT ---
Ok, let's split this
To be exact, the issue here is that libmagic does recognize pcf.gz files correctly *if* MAGIC_COMPRESS flag is used, but the last I tried it in early February caolan said he started getting "cannot create pipe (Too many open files)" errors from magic_file() from within rpm on openoffice.org build.
Didn't have a chance to investigate properly at that time so I just flipped it back to not using MAGIC_COMPRESS which "fixed" the issue.
Looking it a bit more now, it seems that it's .zip files which cause fd leakage in libmagic (4.x and 5.0), other forms of compression seem to be ok. I should have a reproducer case for you shortly..
Created attachment 336274 [details]
Plug fd leak from pipe()
Ok, the issue is easily reproducable with just file(1) itself, any .zip file seems to do and run "valgrind --track-fds=yes file -z <zipfile>".
[pmatilai@localhost ~]$ valgrind --track-fds=yes file -z /usr/lib64/openoffice.org/basis3.0/share/config/images.zip
==11616== Memcheck, a memory error detector.
==11616== Copyright (C) 2002-2008, and GNU GPL'd, by Julian Seward et al.
==11616== Using LibVEX rev 1884, a library for dynamic binary translation.
==11616== Copyright (C) 2004-2008, and GNU GPL'd, by OpenWorks LLP.
==11616== Using valgrind-3.4.1, a dynamic binary instrumentation framework.
==11616== Copyright (C) 2000-2008, and GNU GPL'd, by Julian Seward et al.
==11616== For more details, rerun with: -v
/usr/lib64/openoffice.org/basis3.0/share/config/images.zip: PNG image, 16 x 16, 8-bit/color RGBA, non-interlaced (Zip archive data, at least v2.0 to extract)
==11616== FILE DESCRIPTORS: 4 open at exit.
==11616== Open file descriptor 4:
==11616== at 0x3E20AD8BD7: pipe (in /lib64/libc-2.9.so)
==11616== by 0x4C2F882: file_zmagic (compress.c:383)
==11616== by 0x4C34DA3: file_buffer (funcs.c:206)
==11616== by 0x4C272C6: file_or_fd (magic.c:357)
==11616== by 0x401228: process (file.c:431)
==11616== by 0x401CDB: main (file.c:343)
The attached one-liner seems to be sufficient to fix, but check by yourself, I'm recovering from a nasty stomach flu and not feeling exactly bright and sharp atm...
hello Panu, thanks for the patch, I checked it and it seems to work; if I add it to the next rawhide build of libmagic, can I close this bug as RAWHIDE?
Just toss this back to rpm once libmagic is fixed, I'll still need to enable the compress bit on rpm side of things before this is resolved.
OK, the patch is added in file-5.00-5.fc11, so I'll reassign this...
Okay, looking into compressed files re-enabled in rpm-4.7.0-0.beta1.6.fc11.
Thanks Daniel for fixing these so quickly.