Subject: Re: atime mtime ctime
To: Takahiro Kambe <taca@sky.yamashina.kyoto.jp>
From: Robert Elz <kre@munnari.OZ.AU>
List: netbsd-users
Date: 11/22/2000 21:55:37
    Date:        Wed, 22 Nov 2000 23:06:18 +0900
    From:        Takahiro Kambe <taca@sky.yamashina.kyoto.jp>
    Message-ID:  <200011221406.XAA13666@edge.sky.yamashina.kyoto.jp>

  | > And what about link()?  That, I think, is the killer -- the times are 
  | It's not.  Although link() create another path name for a file,

I think Steve knows that.. but this is really the crux of the issue.

Most people are likely to treat "creation" as being when the (first)
name is assigned to a file, though that turns out to not be very useful
for anything at all.

On the other hand, if you regard (as you do) creation as being when the
data is created, then you have something that (to be consistent) is
close to unimplementable (and which I believe has never been implemented
by anyone).

It requires that (something) be able to tell the difference between

	sed -e '/NO-SUCH-STRING/d' < file1 > file2

(which simply copies the data unchanged, assuming NO-SUCH-STRING doesn't
exist in the file), and hence should leave the creation date from file2
equal to that which it was on file1, and

	sed -e '/THIS-STRING-EXISTS/d' < file1 > file2

which changes the data, and hence should set a new creation time...

And in case anyone thinks they could teach sed (and everything else)
to be this smart, then consider making it work through a pipe...

Creation time simply isn't a useful concept.

kre