On Wed, 2005-01-05 at 09:16, Matthew Miller wrote:
> On Wed, Jan 05, 2005 at 07:57:36AM -0600, Les Mikesell wrote:
> > The traditional in-band delimiter has worked for 20+ years with
> > correctly written tools that do not mangle the messages.
>
> Sadly, not so. What character to you see at the the front of this next line?
>
> >From is used as the delimiter in mbox.
There is a blank line past the end of the previous messge; the current
message starts with
From
> And it can't tell the difference between From to start a new message and
> >From because I happened to start a line that way. And worse, this mangling
> is irreversible -- it's very possible that a quoted line in the body
> contains >From just naturally.
Text has always been subject to various transformations by email
transports. That's one of them. If you need to avoid transformations
there have been various means of encoding available for nearly as
many years as email has existed.
> The alternative -- using "Content-Length", is even more dangerous, because
> it can't be trusted.
Again, mail transports perform transformations. You can't solve
that reliably by changing just one component. You can't trust the
fact that 'your' system works with maildir to continue to work
when the mail is forwarded to or from anywhere else any more than
you could trust "Content-Length", which works equally well in
a closed environment.
> > Unix provides file locking mechanisms that allow different programs
> > to access mailboxes without corrupting them. As usual, if you
> > want to shoot yourself in the foot, it will happily hand you
> > the gun.... Over the years at least 3 different locking
>
> And all my users a gun of their own. All I know is: have a bunch of people
> on a system running pine, and every now and then you'll be fixing corrupt
> mbox mailboxes. You can argue all you want that these people aren't doing it
> right -- but I guess that's the _point_.
My first guess would be that whoever compiled the sendmail local
delivery agent and pine for this system didn't configure the same
locking options. My second guess would be that the mail spool is
on an NFS mount that doesn't have a compatible locking mechanism
with the MTA/MUA's. If none of those are true and the system tools
corrupt messages, your kernel is broken. If people are doing
something like 'vi /var/spool/mail/me' then they deserve to have
problems.
---
Les Mikesell
les@xxxxxxxxxxxxxxxx
