It'd be nice if you could test a build with that tiny workaround and tell me whether it's worth asking for an upload in that state. /proc-based files get some ˙˙no such file or directory˙˙, but others might become useful when it comes to debugging other programs, so we might want to have an half-functional lsof instead of none at all.
Neither "lsof /lib/ld-2.9.so" neither "lsof -i" produces expected output. IMHO, in this case none is better. On the other hand, adapting lsof 4.82 for us should not be too hard: * The March 25, 2009 revision (4.82): corrects an over-zealous test that causes lsof to produce no ouput when the HASSECURITY and HASNOSOCKSECURITTY have been specified at lsof build time; corrects Solaris 10 and above <sys/utsname.h> include/compilation problem and eliminates other Solaris 10 compilation warnings; supports FreeBSD 7.1-PRELEASE; incorporates changes to Darwin lsof, supplied by Apple; enables FreeBSD use of the ${MAKE} variable; improves Solaris VxFS library location test; updates Solaris 10 ZFS support to ZFS pool version 10; updates interpretation of Solaris 10 device numbers; adds rudimentary Solaris 10 sharedfs support; fixes a bug in Solaris 10 zone handling; adapts to changes in FreeBSD 8.0-CURRENT device number computation; selects correct Solaris VxFS library when configuring for gcc; adapts to loss of FreeBSD KAME IPv6 accommodations; adapts to FreeBSD 7.2. Petr