Isn't the solver algorithm too complicated?

[faulesocke]

Wouldn't it be easier to just generate a regular mine field (with all the usual tricks of course), where all mines have pre-determined positions and then implement your additional rules on top of that? For example, if the player clicks on an empty field but he can not know yet that it's empty he looses.

This should work since the set of all non-loosing moves in your game is always a subset of the set of all non-loosing moves in classic minesweeper.

With this you can abandon the SAT solver and all that complicated logic around it.

[pwmarcz]

With your proposal, wouldn't I need to know what are the "fields the player can/cannot know are empty"? And to determine that, I need some kind of reasoning engine anyway, even if it's brute-forcing all the possibilities. So it wouldn't be so different.

Additionally, there is one interesting implication of the current setup: in the case of guessing, the player is able to influence the game somewhat (by choosing a field and forcing it to be empty). This is what makes it truly different from a normal "pre-determined" minesweeper. There is no hidden state anymore, and you could argue the game becomes a complete information game (with human and computer as players), like chess or go.

I think the main source of complication is 1. treating boundary and non-boundary differently, 2. caching past results. These are performance optimizations, and perhaps with a different algorithm they would not be necessary.

[chdoc]

Additionally, there is one interesting implication of the current setup: in the case of guessing, the player is able to influence the game somewhat (by choosing a field and forcing it to be empty). This is what makes it truly different from a normal "pre-determined" minesweeper. There is no hidden state anymore, and you could argue the game becomes a complete information game (with human and computer as players), like chess or go.

Indeed, this is what currently motivates me to come back to this game again and again: I'm trying to understand what's a good strategy for forcing empty fields. Due to the complete information, this is actually more a psychology question than an algorithms question. I find the SAT solver to be much more reliable than me when doing "parity" arguments along long boundary regions. So I try to force fields that allow me to subsequently apply simple reasoning steps.