Suppose you're working in a language with variable length arrays (e.g. with A[i]
for all i
in 1..A.length
) and have to write a routine that takes n
(n : 1..8
) variable length arrays of items in a variable length array of length n
, and needs to call a procedure with every possible length n
array of items where the first is chosen from the first array, the second is chosen from the second array, and so forth.
If you want something concrete to visualize, imagine that your routine has to take data like:
[ [ 'top hat', 'bowler', 'derby' ], [ 'bow tie', 'cravat', 'ascot', 'bolo'] ... ['jackboots','galoshes','sneakers','slippers']]
and make the following procedure calls (in any order):
try_on ['top hat', 'bow tie', ... 'jackboots']
try_on ['top hat', 'bow tie', ... 'galoshes']
:
try_on ['derby','bolo',...'slippers']
This is sometimes called a chinese menu problem, and for fixed n
can be coded quite simply (e.g. for n
= 3, in pseudo code)
procedure register_combination( items : array [1..3] of vararray of An_item)
for each i1 from items[1]
for each i2 from items[2]
for each i3 from items[3]
register( [ii,i2,i3] )
But what if n
can vary, giving a signature like:
procedure register_combination( items : vararray of vararray of An_item)
The code as written contained an ugly case statement, which I replaced with a much simpler solution. But I'm not sure it's the best (and it's surely not the only) way to refactor this.
How would you do it? Clever and surprising are good, but clear and maintainable are better--I'm just passing through this code and don't want to get called back. Concise, clear and clever would be ideal.
Edit: I'll post my solution later today, after others have had a chance to respond.
Teaser: I tried to sell a recursive solution, but they wouldn't go for it, so I had to stick to writing fortran in a HLL.
The answer I went with, posted below.