Can't iterate over all possible values of a table

[duvenaud]

I'm writing a test to see if probabilities sum to one over the set of all possible inputs. Normally, one has to write tedious code to iterate over the entire set, but Dex should shine at this! However, writing e.g.
seqs = for seq:((Fin 3)=>(Fin 4)). seq
crashes the interpreter:

Unexpected type ((Fin 3)=>(Fin 4))
CallStack (from HasCallStack):
  error, called at src/lib/Embed.hs:600:8 in dex-0.1.0.0-WMAsfoEK61B3EtpA608ix:Embed

Amusingly, this somehow corrupted the interpreter which then printed

UnexpectedU nteyxppee c(t(eFdi nt y3p)e= >(((FFiinn  43)))=
>C(aFlilnS t4a)c)k
 C(aflrloSmt aHcaks C(aflrloSmt aHcaks)C:a
l l Setrarcokr),: 
c a lelrerdo ra,t  csarlcl/eldi ba/tE msbrecd/.lhisb:/6E0m0b:e8d .ihns :d6e0x0-:08. 1i.n0 .d0e-xW-M0A.s1f.o0E.K06-1WBM3AEstfpoAE6K0681iBx3:EEtmpbAe6d0
8ix:Embed

[dougalm]

Yes, we don't allow table types as index sets right now. There's no fundamental reason not to, except that the space gets very big very quickly!

The bad error message is for the same reason as yesterday's issue. Since rewrite #101 we're not checking index set type classes.

My guess for the corruption is that we're writing past the end of an array somewhere, which is overwriting the in-memory program. Is it reproducible?

[duvenaud]

Okay, well it's not blocking me. But if it works someday, it'll be a cute example of the power of being able to compose all of Dex's abstractions arbitrarily.

As for reproducing the memory corruption: It happened consistently yesterday but I can't trigger it again today.

[duvenaud]

Closed by #876.