Param match relax

[Graveflo]

#23032

[Graveflo]

paramTypesMatch should only compare operands' type relations to the formal parameter. Currently it calls cmpCandidates which in turn calls checkGeneric and complexDisambiguation on two competing operands. This should not happen because the operands are not a call site and do not have operands of their own, so we cannot preemptively pick the least general, or most complex overload at this time.
That is done during overload resolution because the operand has already been matched to the formal parameters, so the result of checkGeneric and complexDisambiguation are also in relation to the operand. Of course, the parameters to these two procs are formal parameters in this case not operands, thus the isFormal flag.

Scope can still be made useful in paramTypesMatch, so I made that work, but only if the compiler is in need of a resolved symbol (so not for typed and untyped). This is helpful as it allows for simple expressions like:

let x: proc = cmp

If cmp is defined in the same module as x (with no overloads in the module), the user probably wants to select the cmp overload that is in scope. For clarity, this kind of syntax may currently work, but only if the overload that the user wants to select happens to be the least general, or most complex overload globally. In this case it makes more sense to simply be more specific in the let statement.

I disabled datamancer for this PR (again) because after many hours of digging around I cannot fault the compiler for the CI failure. Something is making datamancer's macros pick the wrong type for their df view iterator. This is the problematic code FYI:

import datamancer, seqmath
type
  Foo = object
    fd: float 
let a = [1, 2, 3]
var fm = Foo(fd: 5.2)
let fn = f{ idx("a") > fm.fd }

Now that I am convinced that datamancer is the issue here, I am going to keep investigating, but I'll open this PR for review so that we can start scrutinizing it, as it is now.

[Graveflo]

I think I fixed datamancer, and I'm going to run the CI on one more commit with a couple of changes that I think will be beneficial

[Araq]

Sorry, you need to rebase already.

[Graveflo]

wait a minute... what the?

[Graveflo]

Seems git didn't like what I did there, but it should all be squashed anyway I guess.

There is an ugly part of my changes at the end of paramTypesMatch:

if best > -1 and result != nil:
  # only one valid interpretation found:
  markUsed(m.c, arg.info, arg[best].sym)
  onUse(arg.info, arg[best].sym)
  result = paramTypesMatchAux(m, f, arg[best].typ, arg[best], argOrig)

(and some related bits). I kept this code path in there because that's how it was when I got there. It doesn't have to be weird like that if tyTyped and tyUntyped parameters have no need to do this. I figure they may not have to but I wasn't sure enough to go ahead and do it.

[github-actions]

Thanks for your hard work on this PR!
The lines below are statistics of the Nim compiler built from 94f7e96

Hint: mm: orc; opt: speed; options: -d:release
177411 lines; 7.499s; 766.203MiB peakmem