Spurious rejection of polymorphic types at the syntactic top level

[sauclovian-g]

You can't bind polymorphic terms at the syntactic top level:

f <- return concat;

produces

% saw test.saw
[22:19:20.624] Loading file "test.saw"
[22:19:20.625] Stack trace:
Not a monomorphic type: [t.5] -> [t.5] -> [t.5]

It's not clear what the reason for this restriction might be. One might guess it was cargo-culted from Haskell's monomorphism restriction. Based on https://www.haskell.org/onlinereport/haskell2010/haskellch4.html#x10-930004.5.5, there are roughly speaking three reasons for the monomorphism restriction: confusing lazy evaluation behavior (SAWScript doesn't have lazy evaluation, and besides, the behavior in question is perfectly sound); certain ambiguous cases arising from typeclass constraints (SAWScript doesn't have typeclasses); and other ambiguous cases arising from having modules (SAWScript has no module system). Therefore, there's no need for any related constraint in SAWScript.

Unless some other reason appears, this restriction should be removed.

(This point has come up before, e.g. when I was fixing the typechecker, but I can't find an existing issue about it.)

[sauclovian-g]

So the reason is that it's actually a dynamic check in the interpreter for unification variables leaking out of the typechecker, and the fact that this happens for certain top-level constructs is a result of the interpreter's odd implementation choices (in particular see #2158) combined with (because of those choices) having failed to notice the situation when I was fixing up the typechecker a few months back.

So the check is in fact bogus, but covering up a typechecker bug that needs a real solution.

The structural similarity to certain parts of Haskell's monomorphism restriction is a red herring.

[sauclovian-g]

Fixing the typechecker issue requires fixing the way the interpreter interacts with the typechecker (including, but not only, incrementally for the REPL) and that needs to get in line behind a whole bunch of other things like #2158. So it's not so easy after all.