Date: Sun, 10 Jan 1999 23:34:07 -0000 From: Paul Richards <paul@originative.co.uk> To: "'Poul-Henning Kamp'" <phk@critter.freebsd.dk> Cc: committers@FreeBSD.ORG Subject: RE: sysctl descriptions Message-ID: <E40CBF0361C7D111914000C0F0303D108863@OCTOPUS>
next in thread | raw e-mail | index | archive | help
> -----Original Message----- > From: Poul-Henning Kamp [mailto:phk@critter.freebsd.dk] > Sent: Sunday, January 10, 1999 4:43 PM > To: Jordan K. Hubbard > Cc: sthaug@nethelp.no; des@flood.ping.uio.no; > darrenr@reed.wattle.id.au; > committers@FreeBSD.org > Subject: Re: sysctl descriptions > > > In message <50343.915957694@zippy.cdrom.com>, "Jordan K. > Hubbard" writes: > >> On a fairly current system here there are 319 sysctls. If > each of them > >> had an 80 character description, it would increase kernel > size by 25520 > >> bytes. I'm very willing to pay that price, especially if I > could decide > >> at kernel compile time whether to include the descriptions or not. > > > >So am I. How does the rest of core feel, since phk seems to wish to > >push it to a core vote? > > No, I don't want to push it to a -core vote, I merely used our CVS > convention to indicate where DES could take this as the next step > if he wanted to be (more) confrontational. > > For what it is worth, -core has already turned this idea down when > I proposed it (twice I think), and I don't see any signs of changes. > > Compiling seldom used cryptic (if less then 80 char) messages into > the kernel just isn't the right way to do it. > > Let me state that I think the field in the source is the right > place to write the documentation, but that compiling it into the > >loadable< kernel is not a good idea. > > Several reasoned emails have already pointed out some of the problems: > I18n being just one of them. > > The only reason why I havn't yanked all the descriptions from the > kernel long time ago, was that I had the following fallback plan: > > When ELF kernels were a reality, I would make the LINT kernel > compile the descriptions into a separate section, extract it from > the compiled kernel and stick it somewhere for sysctl to pick up. > That way to update the docs you'd build the LINT kernel. Stuff > not in LINT would have to provide their own files. > > (We now have ELF kernels (Nudge nudge, wink wink!). > > As for making a filesystem interface to sysctl, this could be done, > but no real value was seen at the time and it wasn't done. > > What we really need is a registry without the mistakes M$ did in > NT... Hey, that's exactly what I've been thinking for ages (I spend more time than I'd like working with NT systems these days and there are some points that are worth picking out from it). I feared raising it because in a previous mail from you regarding sysctl you specifically say "sysctl will not become a M$ registry" :-) Maybe we could start a discussion on how we think a registry should work if implemented correctly. Could it for instance include device access information :-) (seriously though, it might be the answer). Paul. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E40CBF0361C7D111914000C0F0303D108863>