His Deeds Are Dust » Uncategorized http://hisdeedsaredust.com surveying sub-optimal solutions Wed, 02 May 2012 13:16:04 +0000 en hourly 1 http://wordpress.org/?v=3.0 pbmclean is dead in Fedora releases of Netpbm http://hisdeedsaredust.com/2009/03/pbmclean-is-dead-in-fedora-releases-of-netpbm/ http://hisdeedsaredust.com/2009/03/pbmclean-is-dead-in-fedora-releases-of-netpbm/#comments Tue, 31 Mar 2009 12:38:52 +0000 Paul Flo Williams http://hisdeedsaredust.com/?p=22 I’ve just been bitten by a bug in pbmclean, one of the tools in the Netpbm package of graphics tools, as provided by Fedora. For a long time, I’ve been building Netpbm from sources, but I’ve been trying to cut down the amount of rework I have to do when updating Fedora, so I’ve moved back to RPMs, but my scanning and cleanup toolchain just broke when pbmclean refused to work at all. A quick test confirmed that it fails for the simplest possible image as well:

$ pbmmake 1 1 | pbmclean > /dev/null
pbmclean: EOF / read error reading a one-byte sample

A bug for this in Red Hat Bugzilla but, if anyone else is experiencing this right now, a simple workaround is pull down the source RPM and rebuild the package without netpbm-10.23-security.patch.

This patch seems to incorporate a load of tests for overflowing arithmetic in image sizes but the “fixes” kill pbmclean and pbmlife, because their tests for row numbers have been incorrectly changed, from:

if (row + 1 < rows)

(in pbmclean.c), to

if (row <= rows)

which clearly isn't the same thing at all. The original tests, although odd-looking, are correct because both pbmclean and pbmlife incorporate row read-ahead, because they change pixels based on the properties of their neighbours.

This bug is present in netpbm-progs-10.35.58-1.fc9.i386.

]]> http://hisdeedsaredust.com/2009/03/pbmclean-is-dead-in-fedora-releases-of-netpbm/feed/ 0