fix #16150 improve type mismatch errors

[timotheecour]

fix #16150

after PR:

Error: type mismatch: got 'seq[string]' for 'p("hello")' but expected 'O = object'

with --declaredlocs --listfullpaths:off

t11377.nim(14, 8) Error: type mismatch:
 got 'seq[string]' for 'p("hello")' [sequence]
 but expected 'O = object' [object declared in t11377.nim(6, 5)]

use -d:nimLegacyTypeMismatch for previous behavior.

CI failure unrelated: #16169

[Clyybber]

I think this is too verbose, consider if/block/case expressions. We should fix the column number instead. (They don't start at 0, try echo undefined)

[timotheecour]

@Clyybber

I think this is too verbose, consider if/block/case expressions.

behavior as implemented in this PR is consistent with sigmatch errors which also show the expression on error, after template expansion, and regardless of how verbose that can become:

when true:
  proc fn(a: int) = discard
  proc bar(a: int): auto = 1.0
  var z = 2
  template foo(): untyped = bar(1+z)
  # fn(foo()) # will show an error that contains: but expression 'bar(1 + z)' is of type: float64
  var a: int = foo() # will now show an error containing:  got 'float64' for 'bar(1 + z)' but expected 'int'

• future work can add an option to instead show the expression as it appears in source code (before template expansion), and apply this to all error messages (including type mismatch and sigmatch errors, in particular the rendering of condition in doAssert failure messages) instead of showing the expression after template expansion.

This is not trivial to do though (related: issues like #7039 and #11789 for const folding or this: #14863 (comment)

Simply remembering both variants can create a compiler that uses an expontential amount of memory. (Nest these calls to get 2x2x2x2x... memory consumption.) There should only be a single AST that works with all the constructs. My "semtityped" AST idea.

There are ways to fix it though.

We should fix the column number instead. (They don't start at 0, try echo undefined)

• That should also be fixed, it's an orthogonal issue