On Thu, Nov 15, 2007 at 09:27:52AM -0800, Bart Smaalders wrote: > Note that adopting Gentoo's scheme would not solve your problem; > as far as I can tell 0.9.8g is not a Gentoo-legal version specifier,
No, it is a valid Gentoo (and Debian) version specifier. Both allow a character at the end of the dot sequence which is compared by ASCII value if the dot sequences are identical. This should be a trivial (and backward compatible) addition to the present IPS versioning scheme. The basic Gentoo format is "version[_suffix[XX]]", where version is a dot sequence followed by an optional letter and suffix is one of "alpha", "beta", "pre", "rc" and "p" followed by an optional integer. For example, 0.9.8g_pre10 is a valid Gentoo version specifier. The hierarchy is like this: alpha < beta < pre < rc < (no suffix) < p One way of adapting this to IPS would result in package names like this: [EMAIL PROTECTED],rc7 This would mean a non-backward-compatible change in the build specifier which would no longer be a dot sequence, but I guess we are early enough in the cycle to be able to accommodate this. Personally, I find this way of representing builds more human-parsable. > If you want the packaging > system to be able to do anything else with the version numbering other > than returning the exact one you asked for, you must impose structure/ > meaning on the versioning that is the same across packages as writing > custom version comparison functions for each package versioning scheme > isn't going to work - especially since they often change over time. Oh, definitely. That way lies madness. But the Gentoo scheme does impose structure and a hierarchy that is the same across packages. I keep harping on the Gentoo/Debian way of handling this because they have had to deal with the same issues we are trying to, and both have come up with feasible alternatives. I prefer the Gentoo one because it is a little more structured (too structured, maybe?) and it accomodates all mainstream packages and pretty much most of the others too. > In addition, package versioning often doesn't reflect compatibility > issues accurately, OpenSSL being a prime example. By assigning our > own version strings we can correct such difficulties. Sure. Cases where we've forked the code significantly, we can always come up with a version which fits our scheme better. I'm more concerned with packages where we'd be using vanilla code, especially given that IPS is a generic packaging system and not an OpenSolaris-specific one. Venky. _______________________________________________ pkg-discuss mailing list pkg-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
