I would say it has not been slow in the last few distros. I really like the delta downloads (it only downloads the changes and not full programs). It is generally a very nice distro to work with.
It seems to me that the delta thing has made updates slower, not faster. I have a reasonably fast network connection and a P4 CPU; maybe with a slow connection and a fast CPU the equation works out better with deltas. But I imagine the real advantage of using deltas is server-side, not client-side.
Deltas or no deltas, apt feels significantly faster, but yum isn't slow enough to be a real problem for me.
It's still slower than apt. Partly this is the use of Python (an extremely unsuitable choice for this), and partly because the repository metadata is less space-efficient (stored as XML and SQLite database dumps). It did get a bit faster around Fedora 10, so it's not painful.
The real problem with yum (and apt) is poor dependency solving. This is an NP-complete problem sufficiently close to 3-SAT that a good SAT-solver can solve this way faster than the stupid Python/C++ heuristics in yum/apt. I wish they'd switch to using something like zypper.
Do you have any benchmarks? Gentoo's Portage is also Python, and it always seemed much faster than apt or yum. Sure, you never want to search for something with it (use eix, which exists for the same reason "locate" exists instead of "find"), but since the bottleneck seems to be more on the speed of the hard drive it doesn't seem like it matters much what language the package manager is in. Plus part of Portage's speed is concurrent downloads as well as allowing concurrent emerges running.
After getting into an argument with one of the yum developers some time ago, I did sit down and attempt to benchmark apt vs yum, but it's impossible to do this. Fedora and Debian are completely different distros, so benchmarking two package installs isn't fair -- the package you are installing might be configured completely differently or have more dependencies on one distro than the other, so it's just not a fair test.
Nevertheless, yum does feel slower, and anyone who uses both daily like I do would tend to agree.
Does Portage do full dependency resolution? That's what kills package managers (see my posting above yours).
I don't see why HN has to switch to feelings all of a sudden, just for this article. Benchmarks are what count.
> so it's just not a fair test.
There are statistical methods for counteracting that, but I don't see the point. It would be non-impossible to construct a test where you tested the overhead of the package management system.
