Re: IMMUTABLE and APPEND-ONLY rationales

[Jamie Lokier]

Andi Kleen wrote:
> >   cp -al kernel kernel-branch
> >   chattr -R +i kernel
> > 
> > Proceed to edit files in `kernel-branch'-- safely knowing that nothing
> > you do there will alter the files in `kernel'.  (Otherwise commands like
> > `prog > file' and some editors will change the hard-linked files).
> 
> Attributes are per inode, not per link.

As they should be.
The point is editing operations have these basic behaviours for "save":

  (1) Simply overwrite the file.  This breaks `cp -al'.

  (2) Copy the original file to make a backup, then write the new
      file over the old one.  Also breaks `cp -al'.  This is
      useful on occasion -- such as when you're editing scripts
      that have multiple names.

  (3) Rename or delete the original file, then create and write a new
      one.  This works nicely with `cp -al'.

> Also you could always archive the same mit chmod u-w

I don't like that because it significantly changes the original tree.
IMMUTABLE OTOH is invisible to most tools.  But it's a fair point.

> (editors usually do > not force writing unless specifically asked --
> and the evil editors that do not would soon learn how to remove
> immutable..)

We can only hope they don't start doing that :-)  It would be much nicer
if they remain ignorant of IMMUTABLE, but switch to strategy (3) for
saving when strategy (2) fails.

-- Jamie