Joker solver crashes

[edderiofer]

Just thought I should document this bug here. It's been known for a while now, but I only noticed last week that it's also affecting the site itself (not just my local testing instance).

This bug affects just about every variant that uses a joker in some way:

• American
• Vietnamese
• Bloody 30-Faan Jokers
• Malaysian
• Sakicards
• and probably a bunch more.

Specifically, whenever a player would win with a joker in hand, the joker solver is called. (Note that it is perfectly possible for a player to win in these variants without jokers.) However, during this process, the joker solver often crashes. Sometimes, it's because the joker solver appears to be fed a nil tile (not sure why, haven't been able to reliably reproduce this version of the error). Other times, the joker solver crashes right at the final get-value step.

This appears to be a bug (or multiple bugs?) in the Elixir parts of the code (more specifically, probably in smt.ex), and my knowledge of Elixir is insufficient to further diagnose the issue or fix it.

---

Weirdly, this bug does NOT affect the following variant(s), which do use jokers:

• Riichi with aka/ao/kindora (the dora tiles are aliased to their regular counterparts, but this doesn't appear to call the joker solver)

and it affects the following variants, but ONLY in my local testing instance and not on the site itself:

• Galaxy Mahjong
• Shouhai
• Golden Chun (the Golden Chun is either a standard chun, or it's a joker for any 5)

[edderiofer]

This bug has apparently been fixed on the website and on my Codespaces testing instance (though if there are lots of jokers, it may take a while), thanks to commit 2f78484.

However, I'm still getting this bug in my local testing instance (on Windows).

Digging deeper with @EpicOrange's help, this might be because ex_smt doesn't seem to have support for Windows, but it's not yet fully clear where exactly things are going wrong and how fixable it is.

[edderiofer]

It appears that ex_smt relies on porcelain, which does not appear to work on Windows.

With the recent changes (i.e. switching from porcelain to ex_cmd in commit bbe35f6; thanks Dani!), along with PR #162, I'm hoping this will be resolved on Windows. Probably going to keep this issue open for a bit longer after the PR is accepted, just to make doubly sure that the bugs are ironed out.

[edderiofer]

I am still occasionally getting errors where the solver is fed a nil tile. This seems to happen most often in Malaysian, but I haven't yet figured out what's causing it.