2013-10-22



The group decided to change seats once a week. It's always diffi-
cult  to  find  a  new distribution that satisfies all wishes. To
provide an alternative approach, I wrote a small programm to gen-
erate  random  distributions.  This  way,  the decision, who sits
where and next to whom, would be transferred to the neutral  com-
puter.  The  complication  was,  that  some people wanted to keep
their seat, for some reasons. Hence, only a subset of  us  needed
to be distributed to a subset of the seats.

First, I solved the problem by shuffling (`sort -R') the list  of
people who want to move and then putting them, one after the oth-
er, onto the empty seats (with awk).   This  was  a  quick  hack,
written in a few minutes.  [0]

Then two of the group told me, that they would also like to  move
to  other  seats  but not to all of them, just to a subset of the
available seats. I thought, that was a pretty fair wish and  thus
my program should support it.  In consequence, however, the naive
approach wasn't sufficient anymore. I started over again.  [1]

I decided to stick to awk to solve the problem.  The  new  script
reads  in a list of names and the seats they would accept. (Those
who don't want to move have only one seat  mentioned.  Those  who
don't  care have `*' as a shortcut for the full list.) The script
reads this data. It tries to fill the names into a list of avail-
able seats. Those who accept only one seat are put there at once.
The others are randomly distributed afterwards. First  those  who
accept the least number of seats, later those who are more flexi-
ble. The random distribution is dumb. The program just  tries  to
put them somewhere, assuming, that this approach will succeed. No
backtracking is done, because  I  thought  it  wasn't  worth  the
costs.  I  consider this decision appropriate as the distribution
will succeed almost always. If not I'd just  re-run  the  script.
This  will  waste  less  time than writing the backtracking stuff
(correctly!).

The script generates pic(1)-output, which is then  rendered  with
troff(1)  to  PostScript  and  then converted to PDF. The textual
representation is done with `pdftotext -layout' because that pro-
vided  better  results  than  generating  text  right  away  with
nroff(1).

[0]  http://marmaro.de/prog/various/seating-plan/version1/
[1]  http://marmaro.de/prog/various/seating-plan/version2/


http://marmaro.de/lue/        markus schnalke <meillo@marmaro.de>