[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: uses for "efficient computation" in various corners of debian?



On Monday 30 October 2006 14:31, sean finney wrote:

> - anything remotely resembling a graph theory problem (pkg deps? bts?)

I guess conflict resolution in the pkg deps graph, 
respecting /etc/apt/preferences *and* the user's wishes would be worth a 
try.  aptitude's algorithm is good, but has one silly bug: when I want to 
install a certain (version of a certain) package foo and there are some 
kinds of conflict, aptitude often tells me that there are conflicts, and 
the typical proposed "solution" is not to install foo.  (same for 
upgrade/uninstall/downgrade of foo)

Also I would prefer the algorithm to try harder to satisfy dependencies by 
upgrading (or perhaps even downgrading) packages instead of by uninstalling 
them.

Yes, I do mix oldstable/stable/testing/unstable + backports + volatile in a 
quite aggressive way, and because I usually know what I'm doing (and 
because the dependencies as such are really quite good), I rarely break my 
system.  Nonetheless, installing app-from-sid-with-many-deps is much harder 
than it ought to be and means a 10-minute session in aptitude, mostly 
hitting +, b, h, enter and q in some random order.

> - introducing new and non-trivial algorithms in bottlenecks anywhere
>   in the "project architecture" (vague, i know)

Interested in sociology?  How about statistics?  cluster analysis of mailing 
list debates, identifying all the various groups of people in the project 
and how they shifted over time, use mailing list archives as input.  That'd 
be fun to read, and would certainly spark at least 6 months worth of 
flamewars rehashing all the old flamewars that caused and killed those 
groups.

Acutal benefit to the project: close to zero, I know.

cheers & good luck
-- vbi

-- 
featured link: http://fortytwo.ch/smtp

Attachment: pgpIUKEAnO5U1.pgp
Description: PGP signature


Reply to: