[PLUG] corruption, but which one is good?

Russell Senior seniorr at aracnet.com
Sat May 21 06:14:54 UTC 2005


>>>>> "Russell" == Russell Senior <seniorr at aracnet.com> writes:

Russell> Following up on the shell question I had earlier today, I
Russell> have a set of 184 files (from Samba server rsync snapshots).
Russell> They are .doc, .xls, .ppt, and .gz files with identical
Russell> filesize and timestamps.  [...]

Er, actually, no .doc files: .xls (162) and .ppt (12) primarily (also
one each .wmf, .bmp and .jnb, a scientific graphing program, the rest
are .gz).

Which is a shame, since with a .doc I'd stand some chance of looking
at a hexdump and seeing text which might indicate which batch is the
corrupt one.

Russell> [...] I detected that the files differ earlier, and the shell
Russell> hackery allowed me to identify the differing bytes.  Since
Russell> most of those are proprietary binary formats, it isn't
Russell> obvious how to figure out which version is correct.  I
Russell> thought I was on to something when I tried gzip -t on each
Russell> version of the .gz files, but both versions test OK.  Gak!

Okay, here's something slightly curious: gzip files same size, same
date .gz files differ but gunzip'd files *do not* differ.

Russell> Any other ideas?

Ah.  First real clue.  This is a Debian/testing box.  One of the .gz
files with differences is a manfile /usr/share/man/man1/mattrib.1.gz.
I compared md5sums with another Debian/testing box and the live
version agrees and my snapshot disagrees.  Checking three other
available system .gz files I find the same thing.


-- 
Russell Senior         ``I have nine fingers; you have ten.''
seniorr at aracnet.com



More information about the PLUG mailing list