Regression: 0.18 code with mapIt() fails to compile on 0.19

[niv]

import strutils, sequtils

let inp = "a:b,c:d"

echo inp.split(",").mapIt(it.split(":"))

This used to work on 0.18 (output: @[@["a", "b"], @["c", "d"]])

0.19 fails with:

test.nim(5, 20) template/generic instantiation from here
../.choosenim/toolchains/nim-0.19.0/lib/pure/collections/sequtils.nim(690, 19) Error: type mismatch: got <seq[string]> but expected 'string = Alias'

[kaushalmodi]

Calling all the recent authors of mapIt:

• @timotheecour ed9fd2b
• @LemonBoy 32b6209
• @skilchen 9bacdd6

[LemonBoy]

Not much of a regression here, for some reason iterators are preferred to procedures in type() expressions (#8901 comes to mind) and that's enough to confuse our simple-minded mapIt.

In the following snippet:

  type outType = type((
    block:
      var it{.inject.}: type(items(s));
      op))

Where op is it.split(":") the compiler picks the split iterator that returns string instead of the split procedure that returns seq[string].

[skilchen]

I only fixed a typo in in this procedure. Personally I would never use it.
You could try:

echo inp.split(",").mapIt(seq[string], it.split(":"))

to get the previous result, or

echo inp.split(",").mapIt(string, it.split(":"))

to get some kind of flatmap...

[kaushalmodi]

@skilchen what is the mapIt(type, arg) syntax?

Update: Answering to myself: https://nim-lang.org/docs/sequtils.html#mapIt.t,untyped,untyped,untyped. Though, this has been deprecated for a while now.

---

About "flatmap", that's just a string representation of the sea, right? (Would be a pain to parse that I think).

This worked when I tried:

echo inp.split(",").mapIt($it.split(":"))

But the output of that won't be very useful.