Sokolution version 6.34 and Sokolution 7.0 have been published

Florent Diedler has released new versions of Sokolution - one of the best Sokoban solvers.

Version 6.34 fixes some bugs compared to version 6.33

Version 7.0 is a special version that only searches for push optimal solutions.

Change log for version 6.34:

• Fix a rare memory corruption
• Fix a bug with symmetry (Thanks Matthias for reporting this bug)
• Fix a bug in diagonal deadlock

Release notes for version 7.0:

• Use a powerful penalty engine for trying to find optimal solutions
• Add a control to check if the solution is really optimal according to solutions found on Sokobano.de website. This control works only for XSokoban, Microban, Microban II and Sasquatch level sets
• Fix the parameter "nbCores" to limit the CPU load used by Sokolution
• Add informations for the user for the generation of the MPDB and MPPDB that could be long
• Misc improvements

You can find details about which levels have been solved in the Sokoban Wiki.

YSokoban 1.850 has been published

George Petrov has published YSokoban 1.850.

The release notes for this new version:

• Bugfix: rare crash during export of solutions.
  

A feature comparison of the main Sokoban programs can be found in the Sokoban Wiki.

Sokoban YASC 1.677 has been published

Brian Damgaard has published a new version of his great and widely used Sokoban YASC program.

The release notes for this new version:

• The "Find similar puzzles" variant of the puzzle duplicate finder has been significantly improved.
  It now uses the Levenshtein edit distance algorithm for comparisons. ("Levenshtein" is spelled correctly - it's not a typo.) 
  Previously, this variant relied on a simpler, more direct comparison method. 
  While it also used normalized puzzle boards and also allowed for a match threshold percentage to account for variations, the results were less flexible than those achieved with the new algorithm. 
  The other two duplicate finder variants, "Find all duplicate puzzles" and "Find duplicates of the current collection", remain unchanged. 
  They still search for exact duplicates (considering puzzle normalization) rather than  similar puzzles, as the computational cost of a similarity-based search would be prohibitive.
• The built-in puzzle generator has been enhanced to produce better-quality puzzles. 
  Previously, it was prone to carving dead-end tunnels and placing goals or boxes in them.
  While this was an easy way for the generator to increase the number of pushes - a key metric for creating more complex and interesting puzzles - such puzzles often turned out to be trivial and too easy to solve.
  For example, here is an auto-generated puzzle with a promising beginning, but after unwinding its initial knot, the remainder becomes trivial:
  "--9#|--#3-4.#|--#--6#|--#3-#|3#-$-#|#-$--4#|#7-#|4#-$$-#|3-##@-##|4-4#".
  The improved algorithm now avoids creating dead-end tunnels in most cases, resulting in more challenging puzzles.
• Bugfix: Filling the optimizer task queue with solutions from a puzzle file containing thousands of puzzles can take so long time that the user might choose to switch to another application during the import.
  This worked as expected, but if the user switched to the Windows desktop instead of another application, then a harmless but highly technical error message would appear on the screen.
  After dismissing the message, the user could restart the import from where it left off.
  This issue has now been resolved for the optimizer task queue and the puzzle generator task queue. 
  The solver task queue, which uses a similar import mechanism, already handled this scenario correctly.
  

• Minor changes.