Re: [BackupPC-users] Could Digest::MD5 be broken on ARM-based computers?

[Jeffrey J. K. <bac...@ko...>]

Jeffrey J. Kosowsky wrote at about 19:19:54 -0500 on Sunday, January 23, 2011:
 > I was testing some of my md5sum routines and I kept getting weird
 > results on ARM-based computers.
 > 
 > Specifically, the pool file md5sum numbers were different depending on
 > whether I computed them under Fedora 12 on an x86 machine vs under
 > Debian Lenny on an ARM-based computer.
 > 
 > This obviously creates issues if you want to move your backup drive
 > between different CPUs.
 > 
 > I narrowed it down to Digest::MD5, by doing the following 1-liner:
 > perl -e 'use Digest::MD5 qw(md5_hex);$file=testfile; $size=(stat($file))[7];$body=`cat $file`; print md5_hex($size,$body) . "\n";'
 > 
 > This should be the same as:
 > perl -e '$file=testfile; $size=(stat($file))[7];$body=`cat $file`; print $size, $body;' | md5sum
 > 
 > For maybe 1% of files in my pool the ARM machine gave the wrong answer
 > when using Digest::MD5
 > 
 > So, something must be wacko in the perl implementation of Digest::MD5
 > on ARM machines!
 > 

Well, what do you know, Perl 5.10.0 (at least in Debian but I think
upstream too) are broken on ARM processors.

Something about 32-bit alignment.
You need to upgrade to 5.10.1 -- and now I wasted a day on this...
And now I need to write code to fix my pool - YUCK!