pub struct Game {
    /// A HashSet with the players waiting to play as account strings.
    lobby: HashSet<String>,
    /// capacity determines  how many people a match contains.
    capacity: u8,
    /// A vector of ongoing matches.
    matches: Vec<Match>,
    /// HashSet indicating for each player which match they are in.
    players: HashMap<String, usize>,
}

I realised that this won't work because if there are 3 matches (0, 1, 2) and I remove 1 because it ends, the players that used to point at 2 will be pointing outside the vector or to an incorrect match.

So I thought the obvious solution was to use a reference to the match: players: HashMap<String, &Match>. But this makes lifetimes very complicated.

What's a good way to deal with a case like these where data are interrelated in the same struct?

all 9 comments

sorted by: hot top controversial new old

[–] expr@programming.dev 7 points 7 months ago (1 children)

Could you not have a hashmap keyed on matches pointing to vectors of strings for the players in each match? Basically modeling the data how you want rather than relying on indexing.

[–] modulus@lemmy.ml 1 points 7 months ago (1 children)

Not sure I understand. What I'm trying to do is something like this:

Poll a stream which takes fedi events. Read player commands.
If an event comes from a known player, check which match they are into.
With that info, get their opponents/coplayers etc and perform the change of state in the game (send replies, next turn, etc).

So what I have as a key is a player name (AP username) and from that I need to find which match they're in.

There's nothing semantically useful about a match ID.

[–] expr@programming.dev 2 points 7 months ago

If you instead made your lobby field a HashMap<String, MatchId>, then whenever you get an event from a known player, you can lookup the match they are in from the lobby map and use that to lookup other data about the match (such as the other players) via the matches field (assuming that was also changed to be a hashmap as described in my previous comment).

[–] QT1@feddit.de 7 points 7 months ago (1 children)

You could store the matches in a HashMap as well, using some MatchId type as the key, i.e., HashMap<MatchId, Match>. Then you can use that as the reference in players: HashMap<String, MatchID>. Only downside is that you have to generate unique MatchIds, e.g., by using some counter.

[–] 5C5C5C@programming.dev 9 points 7 months ago (1 children)

This is exactly the use case that slotmap is meant for. I highly recommend using the library rather than reinventing the concept.

[–] modulus@lemmy.ml 2 points 7 months ago

Ah, that does seem like it will solve the problem. Thanks!

[–] Gobbel2000@feddit.de 5 points 7 months ago (1 children)

I don't think there is a good way of having references within the same struct, but you could store reference counted matches:

matches: Vec<Rc<Match>>,
players: HashMap<String, Rc<Match>>,

You would still have to make sure that the players map is updated, maybe weak references are useful there.

Maybe you could also consider storing the players of a match in the match itself, not outside.

[–] modulus@lemmy.ml 1 points 7 months ago

Thanks, the RC is a possible approach. It seems to violate DRY a bit but maybe there's no way around it.

The reason I had the players outside the match is that I need them there anyway, because when I get a player action I need to check in which match they are, who are their opponent(s) and so on. So even if they're in, they'll have to be out too as there are concurrent matches and the player actions come all through the same network stream.