EWD249 - 78

(either "LONGREP" or "SHORTREP"), however, can be retained. (I mention this as an example of the fact that pearl exchange is by no means restricted to exchange of the bottom pearl.)

With respect to the validity range of concepts along the necklace I would like to ask your attention for an observation which thrilled me the first time I made it. (In retrospection it is pretty obvious and that is exactly why it may be worth-while to be explicit about it.) With each pearl we associate "an independent design decision" and the ordering of the pearls along the necklace therefore implies an ordering of the design decisions. Can we change this order? Yes, we can, although we then have different pearls. By way of experiment I have followed the well-known advice: if you are faced with two primitives -in our case "build" and "print"- decide immediately upon their interface -in our example "image"- so that the two primitives can now be refined independently of each other. So I did, and I came to the following form of necklace

  COMPFIRST
  LINER'
  CLEARFIRST'
  ISCANNER'
  COMPPOS'
  SHORTREP

(the four middle pearls being primed to indicate that they refer to different pearls, although they embody the same decisions as the ones in the original set.) The resulting program is much messier. Why?

Along the necklace we can indicate for each concept its range of validity: of course they overlap and we can view them as the individual threads from which the whole explanation is twined, as a kind of "logical rope". The messy version has a logical rope twined from more and sometimes longer individual threads: its logical rope is thicker, the whole construction is more tightly interconnected. The observation thrilled me because it gave a very convincing demonstration (at least for me!) that elegance, clarity and the like have indeed marked quantitative aspects (as Mozart knew: many if his compositions that make one catch one's breath are misleadingly simple, they seem to be made just out of practically nothing!).