Is it worth making an open source Sudoku library?

JasonLion

I keep thinking about starting a production quality open source library of Sudoku routines. There are many places one can find various bits and pieces of open source Sudoku code, but finding that code takes a fair bit of work, no one source really has all the bits and pieces that you really want, and the details of the licenses vary from place to place. Wouldn't it be nice if there was one place you could look for high quality code that is fast, readable, and fairly complete?

Personally, I want to see something that is in C, C++, or a mixture of the two, has readable well documented code, has a license that allows commercial use, is written so you can use individual parts without bringing in the entire library, and covers a reasonable range of things.

What I don't know is if others would be interested in working on that, or if one of my preferences needs to be changed to gain sufficient support for the project. Some people might be more comfortable with a more restrictive license, others might prefer a different programming language, and so on. The Sudoku field isn't that large, compared to say operating systems or web servers, so a comparatively large percentage of us would need to be on board to have any chance of success.

Is there any interest in such a project, or am I just day dreaming?

humble_programmer · Posted: Sat Jan 16, 2010 5:23 pm Post subject:

I'm on board...I think it's a great idea. Plain ANSI C would probably give the best combination of performance and portability, and I would also suggest that it be hosted on SourceForge, CodePlex, or a similar site. I am willing to contribute code and/or documentation.
_________________
Cheers!
Humble Programmer
,,,^..^,,,
www.humble-programmer.com

ChPicard · Posted: Sat Jan 16, 2010 7:34 pm Post subject: Really good idea JasonLion

Really good idea JasonLion

strmckr · Posted: Sat Jan 16, 2010 11:15 pm Post subject:

id be on board with this idea as well i have all my search routines open sourced already in turbo pascal.. Smile

wapati · Posted: Sun Jan 17, 2010 4:45 am Post subject:

I am an olde and hoary coder.

I think that a library of pseudo code is better.

The comments of well documented code are enough to let
a clever person know what and how to do stuff.

A library of mixed code isn't bad, people that cannot code cannot profit.

Shortest, I prefer pseudo code as it can be read by most.
Good commenting is pseudo code.

I have no code to offer, I wrote at hardware level, around 40 years ago.

Just for a laugh, any of you remember "peeks and pokes"?

gsf · Posted: Sun Jan 17, 2010 4:24 pm Post subject:

a problem with pseudo code is you need a working implementation that faithfully transalates the pseudo code
to prove that it works and is efficient

case in point: what would the pseudo code for bbsudoku look like?
I bet it would be non-trivial to go from that pseudo code back to a comparably fast re-implementation of the original

a problem for real code is agreeing on an api and grid representation
e.g., for algorithmic readability vs. efficiency I strive for readability
but efficiency almost always win out
especially in hot inner loops

I can cite many (non-sudoku) papers that present beautiful and simple algorithms that would never be used in practice

so I think that the purpose of the project must be clear from the start
a tutorial code exposition would be much different from an exposition
that you would want to use in your competitive solver implementation

JasonLion · Posted: Sun Jan 17, 2010 5:35 pm Post subject:

My personal interest is in production quality, readable, well documented, source code. Inner loops in routines that cause user visible delays need to be optimized, occasionally heavily optimized. However, most of the code, which gets run infrequently or at user invisible times, should aim for readability and maintainability, without worrying too much about speed (at least until performance problems become apparent). In practice, that seems to mean very heavy optimization of the brute force solver, and only occasional tweaks to just about everything else.

JasonLion · Posted: Sun Jan 17, 2010 6:59 pm Post subject:

For my first code submission, I offer JSolve, documented in JSolve - yet another really fast brute force Sudoku solver.

This provides a real world example of what I am thinking when I say readable and well documented code.

JSolve was designed to fit the Sudoku routine library category of fastest possible brute force solver.

lkSudoku · Posted: Wed Jan 20, 2010 7:50 pm Post subject:

Personally, I wouldn't like my source code being part of other commercial software, or worse, the assignment of some student who doesn't take the effort of doing his assignments by himself

However, I think that there could be some common C++ interface for all sudoku solvers and generators libraries

It is possible to design an interface that provides all (or most) sudoku related functionality, so that each different contributer can inherit and implement this abstract class with its own implementation

Moreover, once there is a common interface, it is possible to create a timing measurement program that use this interface, so all implementations timings can be compared with the same operations, thus creating a benchmark

In addition, there could be some kind of algorithmic archive for sudoku solvers and generators where only the algorithms behind the implementations are described

wapati · Posted: Fri Jan 22, 2010 12:19 am Post subject:

gsf · Posted: Fri Jan 22, 2010 2:48 am Post subject:

JasonLion · Posted: Fri Jan 22, 2010 3:12 am Post subject:

Right now I am thinking about including things like: a fast brute force solver, a small very easy to port solver that is reasonably fast, puzzle compression, random puzzle generation (with an approximately uniform distribution and various symmetry options).

I wonder about having "human methods" solvers (beyond the simplest methods) and difficulty rating, because I suspect that this is where most of us who have commercial products feel that we have the most value added. If people want to contribute this, then great. But, this isn't something I expect to be included early on.

I do not want to include anything that deals with the OS or GUI. The only exceptions would be a command line utility to allow us to run/test the code and makefiles/project files for the major build systems.

One of the things that most excites me is talking about the various possibilities for the grid data structure with other people who have actual experience with choices other than the ones I made. I have never seen anyone talk about this for more than two or three sentences.

lkSudoku · Posted: Fri Jan 22, 2010 7:06 am Post subject:

strmckr · Posted: Fri Jan 22, 2010 7:47 am Post subject: