Regression of type inference when using templates and a proc with the same name as one of them

[diegomrsantos]

What happened?

The following code doesn't compile

var s: seq[string]

template fail =
  s = @[]

template test(body: untyped) =
  body

proc test(a: string) =
   discard

test:
  fail()

Note that the test proc must have the same name as the test template and at least one parameter in order to reproduce the issue. It compiles If we change s = @[] to s = newSeq[string]().

Nim Version

Nim Compiler Version 1.7.3 [MacOSX: amd64]
Compiled at 2022-11-10
Copyright (c) 2006-2022 by Andreas Rumpf

git hash: 82e99f2
active boot switches: -d:release

Current Standard Output Logs

test.nim(12, 6) template/generic instantiation of `test` from here
test.nim(4, 7) Error: type mismatch: got <array[0..-1, string]>
but expected one of:
proc `@`[IDX, T](a: sink array[IDX, T]): seq[T]
  first type mismatch at position: 1
  required type for a: sink array[0..-1, empty]
  but expression '[]' is of type: array[0..-1, string]

expression: @[]

Expected Standard Output Logs

No response

Possible Solution

No response

Additional Information

It works fine with Nim 1.6.8

[Menduist]

Seems to be introduced by #20091 cc @metagn

[ringabout]

It is the same issue as #20450

[Araq]

Overloading between untyped and string never was supported properly anyway. The real solution is to get rid of untyped as it leads to non-composable DSLs. Easier said than done, I know.

[metagn]

I came across this in #20091 but I thought it was an unrelated bug with calling templates with MCS instead.

Seq type inference currently replaces the [] part of @[] nodes with a typed node, but not the @ part which may be a symbol with a resolved generic parameter (in this case empty). The fix would be to either unresolve this symbol or replace the empty generic parameter. I'm not sure exactly how to do these, hopefully someone can understand my explanation and implement it. Relevant lines (deal with n[0] here):

Nim/compiler/semexprs.nim

Lines 2408 to 2411 in a15872b

	# seq type inference
	var arrayType = newType(tyOpenArray, nextTypeId(c.idgen), expected.owner)
	arrayType.rawAddSon(expected[0])
	n[1] = semExpr(c, n[1], flags, arrayType)