lmb@vicom.com (Larry Blair) (09/25/90)
Bug 1: An admin here reposted everything that he found in `junk' (I stopped creating alt groups long ago) by removing the Message-ID and using `inews -d local'. Unfortunately, our version of C News (PL 18, if you were counting) only inserts `Distribution: local' if there is no Distribution header already present. As a result, any of the repostings that already had a Distribution header got sent out to the world with a new Message-ID. Bug 2: (or maybe not) The path contained in the reposting shows `Path: vsi1!vsi1!...'. I'm pretty sure that 2.11 checked for your site in the path to prevent propagation loops. C News obviously doesn't. Is this a bug (or a feature :-)? The path has to be scanned to determine whether to pass the article or not so why not check to see if your site is already in the list? -- Larry Blair ames!vsi1!lmb lmb@vicom.com
mjr@hussar.dco.dec.com (Marcus J. Ranum) (09/25/90)
In article <1990Sep24.220833.230@vicom.com> lmb@vicom.com (Larry Blair) writes: >Bug 2: (or maybe not) >The path contained in the reposting shows `Path: vsi1!vsi1!...'. I'm pretty >sure that 2.11 checked for your site in the path to prevent propagation loops. I've been seeing exactly the same problem, and was starting to wonder if I was hallucinating. [some of our month old+ news has leaked back out onto the net - thanks to those of you who have pointed this out] I'm running Cnews, patchlevel "most recent that I know of" :) If I post an article with a path like Path: deucac!decuac.dec.com!decuac!decuac.dec.com!mjr it propagagates out just fine. We're also getting (though I hope that's fixed, now) articles making a long, slow loop, and reaching our machine after their history entries have expired here, and going back around. What's the best way to ensure against this kind of problem ? I wonder if various Cnews sites have been lucking out by virtue of Bnews sites "compartmenting" article loops... mjr.
henry@zoo.toronto.edu (Henry Spencer) (09/25/90)
In article <1990Sep24.220833.230@vicom.com> lmb@vicom.com (Larry Blair) writes: >... by removing the Message-ID and using `inews -d local'. >Unfortunately, our version of C News (PL 18, if you were counting) only >inserts `Distribution: local' if there is no Distribution header already >present... I don't think it has ever been clear what the semantics are supposed to be if the same header crops up in both inews options and the article itself. (Actually, we think most of the inews options are silly ideas, since it's no harder to just plug the header into the article.) >The path contained in the reposting shows `Path: vsi1!vsi1!...'. I'm pretty >sure that 2.11 checked for your site in the path to prevent propagation loops. >C News obviously doesn't. Is this a bug (or a feature :-)? The path has to >be scanned to determine whether to pass the article or not so why not check to >see if your site is already in the list? It's one more check to do on a critical path, and one that in normal (and most abnormal) circumstances serves no purpose. Loop breaking is normally done by your neighbors not sending you the article, not by you recognizing that it's yours. -- TCP/IP: handling tomorrow's loads today| Henry Spencer at U of Toronto Zoology OSI: handling yesterday's loads someday| henry@zoo.toronto.edu utzoo!henry
jerry@olivey.olivetti.com (Jerry Aguirre) (09/26/90)
In article <1990Sep24.220833.230@vicom.com> lmb@vicom.com (Larry Blair) writes: >The path contained in the reposting shows `Path: vsi1!vsi1!...'. I'm pretty >sure that 2.11 checked for your site in the path to prevent propagation loops. I don't think that it should. Checking the path to see if a site is already listed is an optimization performed by the SENDER. It is important for UUCP feeds to non-leaf sites. The real protections against duplicates are the message ID and posting date. Articles older than what is retained in the history file should be discarded. I think it is a bad idea to rely to heavily on the path because there is no guarantee against duplicate news names. Before someone mentions the UUCP map project remember that is for UUCP, not news. There are sites that run news without UUCP and sites that just don't register. As a specific example I was sending an aged ihave to sun.com with all the articles from the previous day. I monitored the articles they were requesting (finding a minor bug in my sys line for them) and noticed that they were also requesting articles that they should have received. A little further investigation showed that those articles alread had "sun" in the path so none of sun.com's neighbors were sending them to sun. The problem was that the "sun" in question was "sun.soe.clarkson.edu". So, any articles originating on or passing thru the other "sun" won't get to sun.com. As there is no official control, or even registration, of news sites there is no mechanism to prevent such naming clashes. If sun.com rejected those articles then it was doing the wrong thing as it had never seen them before. I wonder how the small amout of overhead C news saves by not parsing the date compares to the CPU, modem, and PEOPLE overhead spent on processing recirculated expired articles. Jerry Aguirre
lmb@vicom.com (Larry Blair) (09/26/90)
In article <1990Sep25.153457.2612@zoo.toronto.edu> henry@zoo.toronto.edu (Henry Spencer) writes: =In article <1990Sep24.220833.230@vicom.com> lmb@vicom.com (Larry Blair) writes: =>... by removing the Message-ID and using `inews -d local'. =>Unfortunately, our version of C News (PL 18, if you were counting) only =>inserts `Distribution: local' if there is no Distribution header already =>present... = =I don't think it has ever been clear what the semantics are supposed to be =if the same header crops up in both inews options and the article itself. =(Actually, we think most of the inews options are silly ideas, since it's =no harder to just plug the header into the article.) I was so sure of your response, Henry, that I composed the following paragraph yesterday: The decision about what to do when a directive appears as both a command line option and in the file being acted upon is not ambiguous. Unix, as defined by its many utilities, has a clear direction: command line options take precedence. Programming difficulty or execution speed should not be an issue here. As to whether one should use the command line option or not: you can't be serious about your argument. Do you really think that it would have been just as easy to have modified the 400 articles that were being locally reposted as it was to stick `-d local' on the command line? You have made it clear innumerable times that you feel that inews and other 2.11 programs are silly. Why, then, did you implement look-alikes? I suspect it was to make life easier on people making the transition. Unfortunately, users are bound to assume that they operate the same as their 2.11 namesakes, particularly if the documentation doesn't point out any differences. By regarding inews as silly and only making a half hearted attempt to implement it properly, you have unwittingly released hundreds of bogus newgroups to the net and now hundreds of dups. There is nothing wrong with having made mistakes in the original implementation. That is to be expected with a system this size. What is needed is to acknowledge the problems and try to correct them. If you think that inews is silly, yank it. -- Larry Blair ames!vsi1!lmb lmb@vicom.com
mjr@hussar.dco.dec.com (Marcus J. Ranum) (09/26/90)
In article <1990Sep25.210842.9@vicom.com> lmb@vicom.com (Larry Blair) writes: >By regarding inews as silly and only making a half hearted >attempt to implement it properly, you have unwittingly released hundreds of >bogus newgroups to the net and now hundreds of dups. *Henry* did that ?! That seems rather out of character to me. If someone can give me concrete proof, I'll call him and rattle my copy of Sabre-C at him. :) Maybe naming Cnews' "inews" after Bnews' was a mistake, but any news admin who doesn't familiarize h*self with the difference is making a bigger and more damaging mistake. (I know - I made this mistake myself.) mjr.
henry@zoo.toronto.edu (Henry Spencer) (09/28/90)
In article <1990Sep25.210842.9@vicom.com> lmb@vicom.com (Larry Blair) writes: >=I don't think it has ever been clear what the semantics are supposed to be >=if the same header crops up in both inews options and the article itself. > >The decision about what to do when a directive appears as both a command line >option and in the file being acted upon is not ambiguous. Unix, as defined >by its many utilities, has a clear direction: command line options take >precedence... > >You have made it clear innumerable times that you feel that inews and other >2.11 programs are silly. Why, then, did you implement look-alikes? I >suspect it was to make life easier on people making the transition. No, sorry, you have misunderstood badly. We implemented lookalikes because these are part of the *user interface*, and zillions of things would break if we changed that. That does not alter our very low opinion of a lot of the silly, frivolous features that those programs have acquired over time. That being the case, the relevant question is *not* what the "clear direction of Unix" is, but what the existing behavior is. The B News documentation simply does not address the point at all, which can be taken to imply either (a) that no specific behavior is guaranteed, or (b) that this was an oversight in the documentation. In case (a), we're in the clear. In case (b), it's not clear how to proceed, because it is hard to sort out the implementation accidents from the deliberate decisions if you read the code or experiment with it... and we really don't have time for either. If someone would care to find out (not "speculate"; "find out"!) what 2.11 does, we would be interested. (Not "committed to doing the same", mind you, just "interested".) Incidentally, it is unwise to rely on undocumented features. >As to whether one should use the command line option or not: you can't be >serious about your argument. Do you really think that it would have been >just as easy to have modified the 400 articles that were being locally >reposted as it was to stick `-d local' on the command line? Just about, since presumably you did not type "inews -d local ..." 400 times by hand, and you would not do the modifications by hand either. Awk is a very useful program. All of this leaves aside what I'm inclined to consider the real question here: what is the *most useful* behavior? That depends on what you're trying to do with -d. There are about three plausible approaches, given that -d ought to do *something*: 1. -d supplies a default distribution, used only if there is none in the article. This is roughly the effect C News gets now, by accident I think. 2. -d overrides any distribution in the article. This would be the most useful approach in Larry's position when using his solution -- not obviously the best solution to his problem, by the way, but not a ridiculous one either -- but may not be right for others. 3. Presence of both -d and a distribution is an error. This is probably the right solution if you think of inews as primarily a human interface. Contributions from others who have run into this issue would be welcome. -- TCP/IP: handling tomorrow's loads today| Henry Spencer at U of Toronto Zoology OSI: handling yesterday's loads someday| henry@zoo.toronto.edu utzoo!henry