> > > > > - Is it feasible with that much data? > > > > I have 432 packages installed, mostly only one version of them, and my / > > has nearly one million files (/home, /proc, /dev, /sys, /mnt not > > included of course). In view of 15'000 packages, multiple versions of > > each, with multiple use flag combinations for every version I'm rather > > pessimistic, except you have a few hundred dollars on hand and don't > > know what to do with it ;) > > > yeah, after seeing what just my own machine produces (and that's only > including packages where the installing ebuild was still available [I use the > ebuild's md5 to differentiate between non-version-bumped ebuild changes]), I > agree that there is a low likelihood of feasability. > After reading your concerns and adding them to my own, I think I'll just > archive this code and find something else to do with the name "Canoe" (I > really like the joke with Portage :-). Oh well, It was only about a week's > work and 1-2KLOC. Not too big of a loss (no sarcasm). It was fun anyway.
The way I see it, the many permutations or USE flags and archs are overwhelming. OTOH, it would be quite feasible to build such a database for a given arch, profile, and portage snapshot. I would focus on just the GRP packages that come as part of the normal release cycle. There would be no need to collect data from users. BTW, John, you might be interested in my portage patch that adds an "installwatch" feature for pkg_preinst and pkg_postinst. http://bugs.gentoo.org/show_bug.cgi?id=90343 Zac -- [email protected] mailing list
