fix #15836 proc arg return type auto unexpectly match proc with concr… (#21065)

* fix #15836 proc arg return type auto unexpectly match proc with concrete type

* fix #16244

* add test case for #12869

tests/misc/t12869.nim

@@ -0,0 +1,14 @@
+discard """
+  errormsg: "type mismatch: got <bool> but expected 'int'"
+  line: 12
+"""
+
+import sugar
+from algorithm import sorted, SortOrder
+
+let a = 5
+
+proc sorted*[T](a: openArray[T], key: proc(v: T): int, order = SortOrder.Ascending): seq[T] =
+  sorted(a, (x, y) => key(x) < key(y), order)
+
+echo sorted(@[9, 1, 8, 2, 6, 4, 5, 0], (x) => (a - x).abs)

tests/misc/t16244.nim

@@ -0,0 +1,9 @@
+discard """
+  errormsg: "type mismatch: got <int, float64>"
+  line: 9
+"""
+
+proc g(): auto = 1
+proc h(): auto = 1.0
+
+var a = g() + h()

tests/proc/tillegalreturntype.nim

@@ -1,5 +1,5 @@
 discard """
-  cmd: "nim check $file"
+  cmd: "nim check --hints:off $file"
   errormsg: ""
   nimout: '''
 tillegalreturntype.nim(11, 11) Error: return type 'typed' is only valid for macros and templates

tests/types/t15836.nim

@@ -0,0 +1,11 @@
+discard """
+  errormsg: "type mismatch: got <string> but expected 'int'"
+  line: 11
+""" 
+
+proc takesProc[T](x: T, f: proc(x: T): int) =
+    echo f(x) + 2
+
+takesProc(1, proc (a: int): int = 2) # ok, prints 4
+takesProc(1, proc (a: auto): auto = 2) # ok, prints 4
+takesProc(1, proc (a: auto): auto = "uh uh") # prints garbage

tests/types/t15836_2.nim

@@ -0,0 +1,26 @@
+
+discard """
+  action: "compile"
+  disabled: true
+"""
+import std/sugar
+
+type Tensor[T] = object
+  x: T
+
+proc numerical_gradient*[T](input: T, f: (proc(x: T): T), h = T(1e-5)): T {.inline.} =
+  result = default(T)
+
+proc numerical_gradient*[T](input: Tensor[T], f: (proc(x: Tensor[T]): T), h = T(1e-5)): Tensor[T] {.noinit.} =
+  result = default(Tensor[T])
+
+proc conv2d*[T](input: Tensor[T]): Tensor[T] {.inline.} =
+  result = default(Tensor[T])
+
+proc sum*[T](arg: Tensor[T]): T = default(T)
+
+proc sum*[T](arg: Tensor[T], axis: int): Tensor[T] {.noinit.} = default(Tensor[T])
+
+let dinput = Tensor[int](x: 1)
+let target_grad_input = dinput.numerical_gradient(
+    x => conv2d(x).sum())