In other words: an ad hoc, informally-specified, bug-ridden, slow implementation of half of Prolog.
I agree that Prolog would probably fare better as an embedded language like Lua, though - it's best suited to things like rules engines and parsers, rather than whole applications.
Most of these kinds of problems are only about 100 lines of code even in fairly verbose languages, depending on how complicated the set you need to permute / select from is, and how complicated your clauses are.
I speak from experience here, solving packing problems in programming competition questions. Pretty much any high school / college programming competition includes one of these kinds of problems, and the time budget is rarely a lot more than an hour, as it's usually one of the simpler questions.
I guess programming an open-ended system would have taken the 15-year-olds an extra 6-7 minutes.
What a lot of this ends up coming down to is the domain expertises of the code author. If you're uncomfortable with recursion, or writing simple interpreters, then Prolog is a much bigger win. Otherwise, it has a higher hurdle to climb.
Interpreters themselves are pretty trivial. Writing a parser and interpreter for a simple expression language (no control flow or anything like that, just a predicate possibly with some arguments) should by itself take no more than an hour. I know from past experience it takes me about 25 minutes, as it's one of the first things I do when I learn a new language.
Shame on me for being snarky.
Consider that we know, and have known for some time, proper ways to manage dynamic memory. We still consider doing it correctly in practice not trivial and consider automatic garbage collection a productivity win.
This does indeed annoy people who put a lot of work into building complex systems that actually do useful stuff.
This kind of algorithm is normally expected to be written, tested, debugged and presented in one hour in programming competitions with fifteen year-old contestants. And it's only the algorithm that Prolog would help you with in this case; it's the whole application on the exterior where all the real polish goes, where the testing is more subtle. Writing the core algorithm of a packing problem, or a routing problem, is completely trivial compared to the work that goes into the interface for entering the data. Programming competitions work with a text input file to get rid of this complexity. I speak as one who has competed in and won such competitions.
And FWIW, I don't think GC is a productivity win because managing dynamic memory is not trivial. Managing dynamic memory is trivial, but it's tedious, and it warps your program design, forcing it to be more imperative, more mutable and less functional. The cost of manual management isn't in managing the memory per se, but in managing the complexity of an application that can't rely on GC to manage the memory.
IMO GC's productivity win comes from the lowered bookkeeping costs of more modular and expressive programming styles - most notably in the ability to return newly allocated structures as return values from functions without having to negotiate away the ambiguity as to who owns them. It lets you built data structures that may share substructure, and write transformations over them which may or may not rewrite parts of the substructure, and the transformation may or may not be memoized etc. Doing this with GC is straight-forward; handling all these options with manual memory management requires ownership handoff protocols, shared ownership hacks (e.g. reference counts), inflexible conventions, a preference to mutate rather than return new, etc., greatly increasing code complexity.
And I speak of GC as someone who these days normally writes in a non-GC language, but have implemented garbage collectors, both on the compiler (metadata) and runtime (the marking and sweeping) sides.
Would you deploy this code in a production application?
"Programming competitions work with a text input file to get rid of this complexity. I speak as one who has competed in and won such competitions."
So are you then saying that these kinds of problems take about an hour to solve, assuming you are someone who competes in and wins timed programming competitions? Also, what will be the maintenance costs of changing requirements going forward, as compared to adding Prolog clauses?
Having said that, most problems we solve here are NP-Complete, most of the time the trivial pruning you are mentioning does not really extend the limits for the computation.
The knapsack problem, as you know, is NP-Complete.
It's definitely not rocket science. 15 year old kids have been doing this for years in IOI etc.
This is true when the problem is stable and well-specified, but when constraints are being added and removed, when there are "nice-to-have" constraints that save money but may become less tractable as the problem scales, when customers want you to experiment with novel constraints -- you do not want to maintain a homebrew imperative implementation. (I can't say anything about a functional version.) Making your implementation maintainable is tantamount to re-implementing a powerful, general-purpose system such as Prolog.
Where you're dynamically building up facts and clauses, and you need more general solving logic at run time, then I think Prolog is more warranted. But that's a smaller niche, IMO.