overload resolution broken if untyped param shares position with typed param in some overload

[timotheecour]

overload resolution broken if untyped param shares position with typed param in some overload.

Example

template fun(a: bool, body: untyped): untyped = discard
template fun(a: int, body: untyped): untyped = discard
template fun(body: untyped): untyped = discard
fun(true, nonexistant) # ok
fun(1, nonexistant) # ok
fun(nonexistant) # Error: undeclared identifier: 'nonexistant'

Current Output

Error: undeclared identifier: 'nonexistant'

Expected Output

works

Additional Information

devel 1.3.5 0327159

seems like the root cause of those:

• untyped param: wrong error message (undeclared identifier) on type mismatch of another param #12019
• overload resolution (with typed and untyped) doesn't work with iterators #9219
• Exporting a toSeq overload messes up iterator visibility RFCs#512
• Template overloading + identifier injection leads to undeclared identifier #9414
• maybe even block label matching (template with untyped param) or (iterator) treats param as typed, causing CT errors #14778

eg:
#9414 (comment)

This is the documented behaviour, currently overloaded templates must agree on the positioning of untyped parameters. However, there are ideas of how to solve that.

This seems fixable, no?

[Araq]

This is actually covered in the spec. But I agree it sucks.

[timotheecour]

as mentioned in #15940 (comment)

in https://nim-lang.github.io/Nim/manual.html#overloading-resolution-lazy-type-resolution-for-untyped says: "But one has to watch out because other overloads might trigger the argument's resolution:"

in the case of this PR, it's fine since each argument is essentially a disguised typed param; maybe that could be the distinction that makes this overloading sane: when untyped params are disguised typed params (undergo semantic analysis), then overloading can happen even if position of untyped params don't match