try to give a clue if you cannot discard

[fpvandoorn]

Also, if the newest clued card is definitely not playable, then it is treated as a safe hint.

Now players will almost never bomb if they cannot discard. If they cannot discard ...

• ... they will first try to clue 5s if there is no pile with a 4 on top (such clue will be interpreted as a safe hint)
• Otherwise they will clue 1 if all 1s have been played
• Otherwise they will clue a rank which the next player doesn't have in their hand (this search can be extended to suits or other players)
• Otherwise they will play their newest card (which is - I think - slightly better than playing a random card)

---

Note: I noticed one thing in a game which confused me for a while: a play hint can be given by a player, but which is interpreted as a safe hint by the next player (if they clue rank including the oldest unclued card).

[cangiuli]

a play hint can be given by a player, but which is interpreted as a safe hint by the next player

This is definitely a bug. givePlayHint should also require that the hint is not a save hint, preferably by using a modified version of isSaveClue. (I can take a crack at that some time.)

Otherwise they will play their newest card (which is - I think - slightly better than playing a random card)

That seems reasonable.

[cangiuli]

I moved some code around, and defined a type SSet (yeah, I know) for sets of suits, and RSet for sets of ranks. I think these will make some of your code nicer. Soon I will make the clued functions use these instead of lists. (Of course, lists are still needed for ordered information.) But unfortunately these changes will break your new code...

[fpvandoorn]

But unfortunately these changes will break your new code...

That's alright. Feel free to break as much as you need :-)

[cangiuli]

Alright, take a look. I deleted cluedNotRank and cluedNotSuit, and replaced them with possibleRanks and a new version of possibleSuits. Both return sets instead of lists, and they look cleaner. I'm not sure that cluedRank and cluedSuit are still necessary, but I didn't delete them just in case.

In your pull request, I think the only affected function is isNotPlayable, which should get simpler.

[cangiuli]

I should point out that the difference between cluedSuit and possibleSuits is that the latter considers both positive and negative color hints, while the former only considers positive color hints.

[fpvandoorn]

I updated my commit.

In set-list.sml I didn't find an analogue of List.all. To check "for all x in A we have P x" is there a better way than List.all P (ListSet.toList A)?

[fpvandoorn]

Also, is there a way to print out only certain turns in the log, depending on whether some function was called during that turn? That would be helpful for debugging. Imperatively, I would add something like globalVar := true to the function, and let the trace be fn _ => if globalVar then (* print *); globalVar := false else (). Is there a nice way to do this is SML? I suppose we could make play output something of type action * bool, but that is a bit messy.

I'm now working on a better action when you have nothing else to do, but that doesn't happen very often :-)

[fpvandoorn]

Added one more commit.

[cangiuli]

To check "for all x in A we have P x" is there a better way than List.all P (ListSet.toList A)?

This is probably the right way to do it (although it's SSet.toList). If it comes up enough times we can add such a function to UtilHanabi. Note that toList is "free", since an SSet is just a sorted list anyway.

Also, is there a way to print out only certain turns in the log, depending on whether some function was called during that turn? That would be helpful for debugging.

I should factor out some of the printing code so that the players have access to the printing functions (although of course it wouldn't include their own cards). Then you could just print the relevant turns from inside the player logic. A temporary solution is to print "LOOK HERE" in the player whenever the condition comes up, and just search the full game log for it.

[fpvandoorn]

Note that toList is "free"

good point

Then you could just print the relevant turns from inside the player logic.

Oh, that would be nice. It doesn't matter what the current player has to judge whether the move is the correct one.

A temporary solution is to print "LOOK HERE"

Yes, I've indeed been doing that for now.

[fpvandoorn]

I added all my commits to this PR.

[fpvandoorn]

I played around a bit more. I made the following changes:

• if you have a playable card, but the next player needs a save, you save first
• you can give play clues to players other than the next
• if a card is known to be playable (either since a play clue was given, or enough information is known about that card), no play clue will be given to any copy of that card
• I implemented all this in a new player advanced.sml. I removed the memory from simple.sml. The advanced player gets 1-2 more points than the simple player on average.

[cangiuli]

Can't we reconstruct turn number pretty straightforwardly from the rest of the game state? And are we using it anywhere?

[fpvandoorn]

We can reconstruct the turn number from the length of the log. It is currently only used when printing the game state. We can remove it if you want.

[cangiuli]

Even the players can reconstruct it, from the size of the deck / turns left, and number of players. That would be more efficient than checking the length of the log.

[fpvandoorn]

Even the players can reconstruct it, from the size of the deck / turns left, and number of players. That would be more efficient than checking the length of the log.

You mean by counting how much clues have been given? That works most of the time, but not always, since it's possible to play a 5 without getting a clue back.