On Sun, Jan 26, 2014 at 12:48:39PM +0100, David Kalnischkies wrote: > > These patches reduce the startup time of apt-get by ~19% on my system. I > > measured by first doing "apt-get source sl", then running this command: > […] > > Do you have any special reason to pick this command? You mean, for picking "apt-get source sl"? Yes, I use "apt-get source" relatively often and I was afraid of messing around with apt-get code while running it as root – I was doing this on my main box, don't want do accidentially wreck it. > I forgot to remove the binary caches at first, which caused me to see no > meaningful difference, but maybe my machine is just too quick¹. > Removing both caches shows the difference through (for me ~10%, but in a > 5 second run, so this is very close to your results). > I checked with "apt-get check -s" as this just generates the caches (in > memory if run as user) and does the dependency walk which is run by the > other apt-get commands as basic setup. > > (¹ that I would ever be able to say that…) Oh, interesting, after "apt-get check" as root, "apt-get source <name>" becomes much faster. Looks like that cache you're talking about was stale and because I ran the apt-get commands as a user, the cache wasn't updated? Is that case supposed to be possible? > > What do you think? Rewriting the whole source like this would probably make it > > much uglier, but IMO optimizing a few hot code paths this way should be fine. > > Did you tried it with some sort of tool or just by "guessing"? > We had a while back similar improvements with our own tolower > reimplementation, so as long as speedups are observable and are > not to crude I have no problem with adding them. Yes, I used callgrind and kcachegrind to profile it. > (Will follow up on the patches itself in a second) Thanks!
Attachment:
signature.asc
Description: Digital signature