tpf@jdyx.UUCP (Tom Friedel) (08/13/89)
Some assorted C News experiences. Please excuse anything that has already been discussed, or any of my misinformation. I am running 386/ix. The libstdio code fails only in one case, and that is when stdiock.fast -i -u is sent to stdout. I get the message that _cnt is not compatible. Why, and will this be a problem. I am running nn, which expires based upon 3 field in active file. C News does not update this field with expire, and upact is murder on my disk drive. Expire -c 'checks consistency of active file'. What exactly does expire -c do? And what is the best way to use NN with C News? If i get an article that is posted to two newsgroups, one which i get and one which I don't, C News will put it into the first group on the newsgroups: line (creating the directory if necessary), regardless of whether or not this group is in my sys file. When I first tried to build C News with dbz, I got a segmentation fault in dbz hash(). I rebuilt with dbm and all was OK. I am guessing that i needed to do a mkhistory before posting to the newly created C News with dbz, but have not confirmed this theory. Thanks, Tom Looking through the code, you see things like 'super-kludge for B News backward compatability' all over the place. Wouldn't it be great if every one would agree C News is 'the' news software, and all this could be ripped out. -- Tom Friedel Public Access Unix BBS tpf@prism.gatech.edu ...gatech!emory!jdyx!tpf (404) 320-7624 (404) 321-5020
geoff@utstat.uucp (Geoff Collyer) (08/14/89)
Just a reminder: "Distribution: usa" only gets to Toronto by a fluke of our nntp connections. Canada is not yet part of the USA, so na or world distribution is appropriate and I have changed the distribution on this followup to na. Tom Friedel: >I am running 386/ix. The libstdio code fails only in one case, >and that is when stdiock.fast -i -u is sent to stdout. I get the >message that _cnt is not compatible. Why, and will this be a problem. We don't know why your vendor has changed the meaning of _cnt, but it doesn't matter; if you get any stdiock failure, you have to use your vendor's stdio. We hope that if they diddled the semantics of _cnt or _ptr that they have also tuned the guts, but we can't promise. >I am running nn, which expires based upon 3 field in active file. >C News does not update this field with expire, and upact is murder >on my disk drive. Expire -c 'checks consistency of active file'. >What exactly does expire -c do? And what is the best way to use NN >with C News? I'll let Henry give the definitive answer on upact (which I only run weekly and I don't think zoo runs it at all). expire -c doesn't remove nor archive any files, it just checks that explist and active are okay, generating error messages (and I assume bad status) if not. expire(8) really does say this. >If i get an article that is posted to two newsgroups, one which i get >and one which I don't, C News will put it into the first group >on the newsgroups: line (creating the directory if necessary), regardless >of whether or not this group is in my sys file. If an article is accepted for filing by relaynews (because your machine's (sys file) subscription list matched at least one newsgroup in the article's Newsgroups: line), relaynews will then attempt to file the article in all the groups in the Newsgroups: line. (We shall ignore the precise rules for junking here; see relaynews(8) if you are interested.) Since the active file (which should really be called "groups", not "active") defines the set of newsgroups accepted locally, it is consulted during filing; it is assumed that if you don't want a group, you either have its flag field set to "x" (meaning don't junk, just ignore) OR you don't have it in your active file and your subscription list refuses it. Leaving a group un-x'ed in your active file even if refused by your subscription list will cause occasional cross-posted articles to be filed in that group, and leaving it permitted by your sys file but not in active may result in the article being junked. >When I first tried to build C News with dbz, I got a segmentation fault >in dbz hash(). I rebuilt with dbm and all was OK. I am guessing that >i needed to do a mkhistory before posting to the newly created C News >with dbz, but have not confirmed this theory. Since dbm and dbz use completely different file formats, you have to pick one and use it consistently everywhere. The core dump in dbz is one reason that we have been a bit slow to pick it up and endorse it, though some time we hope to have an improved dbz which we will endorse. >Looking through the code, you see things like 'super-kludge for B News >backward compatability' all over the place. Wouldn't it be great if >every one would agree C News is 'the' news software, and all this could >be ripped out. The only kludges that come to mind immediately are inews -C, Supersedes:, foo.bar.ctl, Subject: cmsg, (unbatched) ihave/sendme, and "internet" and "backbone" in mailpaths. All but "backbone" are related to control messages, which provide much of the complication in news software anyway. inews -C we plan to deal with in the next patch. Supersedes: only affects comp.mail.maps and can either be handled by expiring comp.mail.maps faster than normal or by running superkludge from crontab. foo.bar.ctl is recommended by RFC 1036 so we chose to support it rather than say "C news conforms to RFC 1036 except for the following kludges: ..."; it is truly ugly and utterly unnecessary, but god only knows how many newsreaders rely upon it. "Subject: cmsg " is even worse, but RFC 1036 requires that it be recognised. Unbatched ihave/sendme is a bad idea and the description of ihave/sendme in RFC 1036 is so ethereal that I don't think it can be read as requiring unbatched ihave/sendme. Batched ihave/sendme isn't a lot better, but we had to provide some form of ihave/sendme to get the 1036 Goodhousekeeping Seal of Approval, it's only three pages of code, and I was waiting for Henry anyway. "internet" doesn't really belong in mailpaths (which should really be called "modroutes"), it was nailed on the side, but this brings up the whole ugly subject of Path: vs From: for replies. "backbone" means "all"; it's just silly syntactic sugar. -- Geoff Collyer utzoo!utstat!geoff, geoff@utstat.toronto.edu
woods@robohack.UUCP (08/14/89)
In article <1989Aug14.040740.3151@utstat.uucp> geoff@utstat.uucp (Geoff Collyer) writes: > Tom Friedel: > >I am running 386/ix. The libstdio code fails only in one case, > >and that is when stdiock.fast -i -u is sent to stdout. I get the > >message that _cnt is not compatible. Why, and will this be a problem. > > We don't know why your vendor has changed the meaning of _cnt, but it > doesn't matter; if you get any stdiock failure, you have to use your > vendor's stdio. We hope that if they diddled the semantics of _cnt or > _ptr that they have also tuned the guts, but we can't promise. Just as a side note in this conversation: I am running beta C News on ISC 386/ix 1.0.6 with the 1.0.5 compiler and Jun23 C News on a 3B2 with SysVr3.1 and CPLU 4.1. I have not done the stdio tests on the 386, but when I did them on the 3B2, I had the same result as Tom. However, since I'd already been running with the libstdio code in both places for >2 weeks on the 3B2, and >6 months on the 386, I've continued to use it. The only odd, un-explicable thing that has happened was an article being tossed into the errlog because relaynews thought it was out of sync (with "relaynews out of sync, tossing:" messages). Since this happened on both machines, I though it might have been a bad batch, though I never got a chance to examine the batch. [Geoff:] > [Tom:] > >When I first tried to build C News with dbz, I got a segmentation fault > >in dbz hash(). I rebuilt with dbm and all was OK. I am guessing that > >i needed to do a mkhistory before posting to the newly created C News > >with dbz, but have not confirmed this theory. > > Since dbm and dbz use completely different file formats, you have to > pick one and use it consistently everywhere. The core dump in dbz is > one reason that we have been a bit slow to pick it up and endorse it, > though some time we hope to have an improved dbz which we will endorse. I use dbz on both machines, and have definite evidence of other successes. Someone should sdb or dbx the sucker! I can't (yet), 'cause I've not had the core dump. :-) I did accidentally run the fake dbm here on the 3B2 ('cause I forgot that the build stuff didn't recommend it), and it ran fine too, (but slow). The one thing I was impressed with was watching mkhistory work on a system with ~50 Kb news in the spool, and using the fake dbm (it was a 25(? fast anyway) MHz AT&T 6386). -- Greg A. Woods woods@{robohack,gate,tmsoft,ontmoh,utgpu,gpu.utcs.Toronto.EDU,utorgpu.BITNET} +1-416-443-1734 [h] +1-416-595-5425 [w] Toronto, Ontario; CANADA
henry@utzoo.uucp (Henry Spencer) (08/14/89)
In article <1989Aug13.161806.15829@jdyx.UUCP> tpf@jdyx.UUCP (Tom Friedel) writes: >... Expire -c 'checks consistency of active file'. >What exactly does expire -c do? Basically, expire -c does all of expire's startup processing -- including reading and parsing the expiry control file and the active file -- and then exits. It's mostly a debugging aid for people fiddling with control files. > And what is the best way to use NN with C News? As we've said before in this forum, news readers that expect a contiguous or even near-contiguous sequence of articles between the "min" and "max" numbers from the active file are broken, pure and simple -- this assumption is just plain wrong. It was fixed in the major news readers, e.g. B News "readnews", years ago. All I can suggest is running "upact" nightly in the wee small hours. >If i get an article that is posted to two newsgroups, one which i get >and one which I don't, C News will put it into the first group >on the newsgroups: line (creating the directory if necessary), regardless >of whether or not this group is in my sys file. As Geoff has pointed out, there are complicated interactions between the sys file and the active file here. We spent considerable effort and head-scratching getting the C News behavior "right", in the sense of being the behavior that was most useful and gave the most control. It's not, unfortunately, trivial to understand. >When I first tried to build C News with dbz, I got a segmentation fault >in dbz hash(). I rebuilt with dbm and all was OK. I am guessing that >i needed to do a mkhistory before posting to the newly created C News >with dbz, but have not confirmed this theory. Correct. Dbm and dbz files are *not* compatible; it has to be all one or all the other. -- V7 /bin/mail source: 554 lines.| Henry Spencer at U of Toronto Zoology 1989 X.400 specs: 2200+ pages. | uunet!attcan!utzoo!henry henry@zoo.toronto.edu
epsilon@wet.UUCP (Eric P. Scott) (08/14/89)
In article <1989Aug14.040740.3151@utstat.uucp> geoff@utstat.uucp (Geoff Collyer) writes: > Supersedes: only >affects comp.mail.maps Supersedes: is used by news.announce.newusers and we very much depend on it! -=EPS=-
davison@drivax.UUCP (Wayne Davison) (08/15/89)
tpf@jdyx.UUCP (Tom Friedel) writes: >I am running nn, which expires based upon 3 field in active file. >C News does not update this field with expire, and upact is murder >on my disk drive. Are you at least running the modified version of upact that replaces the min=`ls $NEWSARTS/$dir | egrep '^[0-9]+$' | sort -nr | tail -1` with min=`lowest $NEWSARTS/$dir` (after building and installing lowest, of course)? This in itself decreased the running time of upact from slightly over 14 minutes to about 7 minutes & 20 seconds. (We run 4.3BSD on a vax 11/780 with only about 30 meg of news on-line). I only discovered this "option" after writing an entirely C version of my own (but it executes in 30 seconds, so I'm not complaining). I'm going to send it to the C news folk for their perusal as an optional upact (it requires directory functions, so they probably won't want to make it the default, like "lowest" isn't the default). Also, it is my firm belief that empty newsgroups should have their min field set to one higher than their max field -- newsreaders like rn & nn interpret this as an empty group rather than a group with 1 non-existant message. I wrote my upact that way, and modified addgroup to start with an initial value of "group.name 0000000000 0000000001". -- Wayne Davison \ /| / /| \/ /| /(_) davison@drivax.UUCP (_)/ |/ /\| / / |/ \ ...!amdahl!drivax!davison
zeeff@b-tech.ann-arbor.mi.us (Jon Zeeff) (08/15/89)
>When I first tried to build C News with dbz, I got a segmentation fault >in dbz hash(). I rebuilt with dbm and all was OK. I am guessing that >i needed to do a mkhistory before posting to the newly created C News >with dbz, but have not confirmed this theory. You are correct. Dbz uses a different format for the .pag file so you need to rebuild it. -- Branch Technology | zeeff@b-tech.ann-arbor.mi.us | Ann Arbor, MI
henry@utzoo.uucp (Henry Spencer) (08/16/89)
In article <24E7016E.1749@drivax.UUCP> davison@drivax.UUCP (Wayne Davison) writes: >... I'm going >to send it to the C news folk for their perusal as an optional upact (it >requires directory functions, so they probably won't want to make it the >default, like "lowest" isn't the default). We generally have a fairly low opinion of upact, but so many people would like it to go faster that I'm willing to consider the idea of a C version. >Also, it is my firm belief that empty newsgroups should have their min field >set to one higher than their max field... Hmmm... Interesting idea. Anybody know of anything that would break? -- V7 /bin/mail source: 554 lines.| Henry Spencer at U of Toronto Zoology 1989 X.400 specs: 2200+ pages. | uunet!attcan!utzoo!henry henry@zoo.toronto.edu
amanda@intercon.uu.net (Amanda Walker) (08/16/89)
In article <1989Aug15.175540.23669@utzoo.uucp>, henry@utzoo.uucp (Henry Spencer) writes: > >Also, it is my firm belief that empty newsgroups should have their min field > >set to one higher than their max field... > > Hmmm... Interesting idea. Anybody know of anything that would break? Not off hand, and I know some code that it would un-break (well, it would avoid calling a routine to check to see if there really is an article or not, which would speed things up a bit). I second the motion! -- Amanda Walker InterCon Systems Corporation -- amanda@intercon.uu.net | ...!uunet!intercon!amanda