Re: bug in du --time=atime

[William Brendling]

On 7/19/05, Paul Eggert <address@hidden> wrote:
> Hmm, why not simply stat the directories before reading them?  That
> would give more useful behavior, and it won't require the complexity
> of an additional option.

That will require me to understand the "tree walking" code (which so
far I have not needed to). According to the notes in du, FTS does a
depth first search, in which case it may have already read the
directory before calling the process_file callback. However there is
also a comment to the effect that process_file is called more than
once for a directory, perhaps the first call is before the directory
is read. I will have to look into it further.

There was a practical reason behind my suggestion of an exclude
directory times option. Suppose that you are using "du --time" to do a
weekly scan for unused directories (which is why I started on the
patch in the first place). Suppose also that you are using the atime
option. Then, at the moment, you will always get the current date
(which is no help). If the stat and read are transposed, then you will
get a date of not more than a week ago (the date of the previous
scan). This is not a lot more useful. What you want to know is that
none of the files in the directory have been accessed for 12 months
(or whatever).