du vs big dirs

[utzoo!henry]

Du(1), as distributed, fails on large directories ( >= 32KB ).  Such
directories are unlikely (and unwieldy) but not impossible, particularly
if you have a mail/news system that creates lots of little files in the
same place.  The fix is to declare du.c/descend()'s variables "dirsize" and
"offset" to be long instead of int.

Also, du has an inherent problem:  it has no way to tell whether a file
has "holes" in it, i.e. whether all the bytes indicated by st_size really
are allocated.  This can result in overreporting of disk consumption,
although this is not a major problem since holes in files are uncommon.
Just as well:  there is no easy fix for this one.