henry@utzoo.UUCP (Henry Spencer) (08/04/83)
People talking to sites with fairly new uucps should be aware that
there has been an important and subtle change in file naming. It
seems to have snuck into the Berkeley stuff around 4.1c, and Peter
Honeyman's latest uucp at allegra seems to do something similar,
so perhaps it arrived in System N (N == 3, 4, 5, ...).
The change is like this. In old uucps, data files being sent to
a remote site ("remote") by my site ("sender") for use by rmail etc.
have names starting with "D.remote". This does involve one slightly
unfortunate possibility, in that if the remote site is receiving from
two people at once, and their sequence numbers happen to be similar,
collisions can happen.
In the new uucps, the data files have names starting with "D.sender".
This is more consistent -- all outgoing files contain the name of
the sender, and since they also have sender-generated sequence numbers
there is no possibility of collision -- but it is different. Moreover,
there is some chance of collision when a new-uucp site sends to an
old-uucp site, because old-uucp sites use names of that form for
stuff that is headed the other way. This chance is fairly small if
the old-uucp site has a really old uucp, because the new uucps also
use the big-alphabet sequence numbers and hence usually have sequence
numbers that an old uucp won't generate, but it's not zero. Beware.
On the whole I think the change is an improvement, but I DO WISH IT
HAD BEEN ANNOUNCED! It is really upsetting to find out about something
like this by observation, and to have to spend some frantic moments
convincing oneself that it isn't going to break anything. It doesn't
break anything here, but other sites may not be so lucky. C'mon guys,
how about letting us know what to expect? Network compatibility is
not an area that benefits from surprise changes.
--
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henryhoney@allegra.UUCP (08/04/83)
the name of the data file is irrelevant -- the C. and X. files specify the names of the data files, so when redman, nowitz and i changed the filenames, we didn't feel compelled to announce it or discuss it (other than a brief spat with lauren -- naturally he won). when i send to you, it's a win for me to use D.myname* since this guarantees no collisions (in both our spool directories). (as an aside, our D. naming convention is so totally bizarre, there's virtually no chance of a collision in an old site's spool directory.) our moments considering whether this would break anything were not frantic, like yours; they were positively serene. peter honeyman ps: this did not "sneak" into 4.1c -- i suspect the 4.1c uucp people had the same notions independently.