[Aptitude-devel] Bug#142656: aptitude: commandline search should be faster

[Manuel A. Fernandez Montecelo]

Control: tags -1 - confirmed
Control: close -1


Hi,

2004-02-03 20:12 Daniel Burrows:
>  Just a note: I investigated this bug a bit recently, and determined
>that almost all the time is spent in startup code -- the search itself,
>while not as fast as it could be, takes up over half the runtime.
>(unscientifically determined by modifying the program to halt
>immediately after initializing itself)
>
>  I would suspect at this point, with almost no evidence ;-), that the
>problem is at least partly the huge text database of states that aptitude
>loads on startup.

I've been doing some optimisation of the initialisation in the 0.7
series which should have helped with this.

On my system (with unstable and experimental, dbgsym packages, and
multi-arch enabled -- more than 90K packages in total), "aptitude search
aptitude" takes just under one second, "aptitude search
'~ddoes-not-exist'" just under two seconds.  "aptitude search '~d.'"
takes 3.6 seconds, and the output is 92+k lines printed.


Comparing aptitude and apt for search speed is apples and oranges,
though.

aptitude does much more than apt:

- can search with patterns,

- match by debtags or user-tags which apt doesn't even support (for
  which it has to read different files --and not very small--), just in
  case that the user wants to use these patterns to match against
  packages

- access additional info like installed state + planned action +
  automatically installed (which shows by default)

- the columns can contain (and by default, they do contain) much more
  information in addition to apt-cache's "name and description only"

- then sort and columnize the result


So I don't think that aptitude's search is specially slow or
inefficient, or that its current speed/slowness is a usability problem,
or that spending significant time [1] to improve the efficiency is a
good use of developer efforts at this point.

So I am closing the report now, after more than 14 years open (and more
than 12 of inactivity) :)


[1] after what I spent already to optimise the 0.7 series, which was
    about a week, I estimate that even very minor gains (<5%) can take
    the best part of a day at the very least


Cheers.
-- 
Manuel A. Fernandez Montecelo <manuel.montezelo at gmail.com>