> "David T. Ashley" <[EMAIL PROTECTED]> writes:
>
> > OK, I've reviewed Util.py.  I can't find any evidence of
> > serialization.  I also reviewed the 'C' fopen() function with "a+"
> > mode, and I also can't find any evidence of serialization.
>
> See http://mla.libertine.org/tmda-workers/200110/msg00031.html
>
> For a more detailed treatment of a+/O_APPEND, I'd recommend Richard
> Stevens' _Advanced Programming in the UNIX Environment_.
>
> > My concern is if two processes try to append to the file at the same
> > time.
>
> Don't be concerned, it's safe on any modern UNIX which follows the
> POSIX standard.
>
> What you don't want to do is hand edit the CONFIRM_APPEND file while
> qmail is running, since there is no locking going on between TMDA and
> your text editor.

I'm still very suspicious.

I looked up Unix file semantics.

I'm not concerned about Unix file semantics (about the OS internals).  I'm
concerned about the way the Python interpreter behaves.

For example, what prevents the Python interpreter from appending a string by
appending it character-by-character?  In that case, the competing "appends"
may intertwine in an unexpected way.

In my mind, this is still an open issue until the source code for Python is
examined.

It was also mentioned in the thread above that nobody has seen such a bug
surface.  There are a lot of reasons possible for this.  The probability of
two confirmation messages coming back simultaneously is very small.  If that
happens, qmail and the OS may "tend" to serialize processing of mail (for
reasons I won't mention).  So, the multiplied probabilities may be very
small.  So I don't take the lack of observation as hard evidence.

In my mind, the only way to close out this issue is to examine the Python
interpreter source code.  I will do that, I believe.

Whatever the outcome, I believe the probability is very small of this
happening in practice.  I'm just paranoid, I think.

Thank you also for the book recommendation.  This is exactly the kind of
book I've been looking for.

Dave.

_____________________________________________
tmda-users mailing list ([EMAIL PROTECTED])
http://tmda.net/lists/listinfo/tmda-users

Reply via email to