mhparker wrote:
Firstly, I would advise against an over-reliance on a specific software solver, which especially applies in the case of JSudoku and advanced vanilla. As enxio27 already pointed out somewhere on this forum, JSudoku was originally designed for solving Killers.
Correct, but vanilla techniques are also applicable to killer. I agree the ordering of some techniques may look strange for vanilla. For example I've put naked pairs and triples before intersections (locked candidates) because I find it makes more sense for killer of average difficulty which are typically solved by first determining combinations for cages yielding naked subsets and then determining the particular location of the digits in the cage. This is quite different from the way one solves vanilla sudokuy of average difficulty, where one typically focus on one digit at a time.
mhparker wrote:
Having said that, Jean-Christophe has clearly been working hard recently to iron out some of its weaknesses in this area (for example, by building in ALS-XZ, Sue de Coq, Finned X-Wing/Swordfish and Uniqueness Test solvers).
To come: 2-string kite, empty rectangle, remote pairs.
mhparker wrote:
But there are still some things I'm asking myself, like:
1.Why JSudoku places Skyscrapers so high in the default priority sequence, even above XY-Wings (for example)? I personally would expect most solvers to try XY-Wings first.
Not everybody will ever agree on a definitive sequence of techniques. I personally find it easier to spot skyscrapers than XY-Wings. Udosuk has no problems spotting naked quints, but find it hard to spot hidden pairs. Børge find 2 string kite easier to spot than skyscraper... That's why you may reorder the techniques to match your personal taste.
mhparker wrote:
2. I note that some Skyscrapers found by JSudoku are based on conjugate pairs all the way round (i.e. strong link used as weak link in the middle). AFAIAA, most vanilla solvers would report this as Simple Coloring.
There is no coloring at all in JSudoku, mainly because the UI was not designed with coloring in mind... JSudoku uses chains made of strong and weak links. AFAIK it yields the same eliminations as coloring, but using another method/paradigm. I could add some coloring, but the more techniques I add, the slower it will get. JSudoku already look too much "bloatware" for my taste...
mhparker wrote:
Lastly, how does one reset the solver sequence to default in JSudoku (I'm using version 1.3b1). I naiively expected to find something like a "Restore defaults" option on the context menu for the solver page.
It's now available in v. 1.3b2. It's in contextual menu for the solver page.