Map key becomes "valid/existing" after value lookup

[luauser32167]

fn main() {
  m := map[int]str{};
  printf("%d\n", int(validkey(m, 69105))); // not valid
  var s: str = m[69105];
  printf("%d\n", int(validkey(m, 69105))); // but now valid
}

[vtereshkov]

@luauser32167 This is the intended behavior, similar to that of std::map in C++. It is discussed in detail in #170

[luauser32167]

This is a bit surprising...
@vtereshkov have you considered the approach that 'stb/ds.h' uses for its hashmaps? It has a setter function which sets the "default value" for the hashmap. When a key is used for lookup with the function 'hmget', and it isn't found, the value of the default item is returned.

Basically it has the following semantics:

  m1 := map[int]str;
  printf("%s\n", m1[69105]); // "" (empty string, because the default value is zero initialized)
  mapsetdefault(m1, "leaves");
  printf("%s\n", m1[69105]); // "leaves"
  printf("%s\n", repr(m1)); // { } (no keys)

  type Point = struct { x, y: int };
  m2 := map[str]Point;
  m2["origin"].x = 10; // sets the default value's 'x' field to 10 (because the key "origin" was not found)
  printf("%s\n", repr(m2)); // { } (no keys)
  m2["origin"] = Point{10, 20};
  printf("%s\n", repr(m2)); // { "origin": Point{x: 10, y: 20} } (now has 1 key called "origin")
  printf("%d\n", mapgetdefalut(m2).x); // the default value's 'x' field is still 10

[vtereshkov]

@luauser32167 The following is even more surprising for me than having m2["origin"] initialized on first access:

m2["origin"].x = 10; // sets the default value's 'x' field to 10 (because the key "origin" was not found)
printf("%s\n", repr(m2)); // { } (no keys)

Assigning m2["origin"].x = 10 is a clear indication that I want m2["origin"] to exist. But m2 is still empty. That's weird.

Here is what C++ is doing: https://www.onlinegdb.com/_2_bEnIeRK.