> So modern Linux package managers do not have any of these qualities ?
Some of them ? Of course. All of them ? No. Just a few points: 1. They usually upgrade a live system (and while it's possible for some to do a non-live upgrade, the live upgrade is exactly how pretty much all of them operate by default). 2. The upgrade process is usually very slow, I've been upgrading Fedora, Ubuntu, OpenSuse, what they have in common ? A very slow and IO bound upgrade, during the "update/installation" phase which took most of the time the whole system slows down, to put into perspective, last time I did a nightly upgrade of fedora it took around 1.5 hours, while the download phase took only 10-15 minutes. 3. The whole package content is downloaded (recently there has been a trend towards moving to downloading deltas only, but still the default in most is to get the whole content). 5. Usually the design is not cross-platformant, IPS can even run on Windows. > Strange!! And how difficult it is to implement no-scripting limitation in an > existing packaging system ... patching something for that would be just as hard as implementing that particular functionality from scratch, and you would have to maintain patches with the upstream (unless you have an ideally de-coupled backend API, which I doubt would be feasible to have once you start adding more and more features), forced to use whatever libraries/languages the upstream uses, and so on. > Like ZFS. The ideas expressed in ZFS are revolutionary > to say the least and could not have been done by re-using existing > stuff. However there should exist a balance between redo everything and > re-use otherwise one would start re-writing every piece in the name > of innovation. While I agree with you about that, I still stand by my point that IPS is a good thing, and given Sun's need for paid-support repositories, integration with zones, SMF support, ZFS support, they would've ended with a lot of patches on their hands. As for the time it has taken, it's not just the development time, once you've written some complex piece of code in say N months, you could easily implement the whole thing from scratch in N/M months, it's the slow development of ideas that should be counted too. Moreover, now when most of the code is in place and working, the team can quickly introduce new features as they are well familiar with the codebase and know the design well to easily extend IPS. > The user side experience of IPS is no doubt very > good but is no different from a good Linux package manager > like Smart/Yum (with the exception of ZFS features). Unfortunately, I don't find yum to be a good package manager for some of the reasons I listed above. > From a developer point of view these qualities could have been got by far > less effort and far less code/complexity. In the short run ? Maybe. In the long run ? I don't think so. -- This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list [email protected]
