> char default is 0xFF....

Invalid values, perhaps? Other than making initialization mandatory, this seems like a reasonable way of catching them. For integer types it's not really possible because all their values are valid.

  import std.sumtype;
  
  void main(){
   struct A { int a; }
   struct B { string b; }
   struct C { bool c; }
   
   alias TaggedUnion = SumType!(A, B, C);
   
   auto tgu = TaggedUnion(B("hi"));
   int i = tgu.match!(
    (A a) => a.a,
    (B b) => b.b.length,
    (C c) => c.c * 5
   );
   assert(i == 2);
  }

I know of sumtype, and i think it is a mistake for D to rely on this for such important language feature

I'm not a language dev, so i can't do much to help, using switch/union/enum is more natural, is cleaner, and is easier to add support for IDEs, everyone on the same page

That's in areas like this where the language falls short, and i can see people instead choosing alternatives

https://github.com/dlang/phobos/blob/master/std/sumtype.d

a mess of templates and imports to std.traits

using this will make your compile speed tank.. another reason to avoid

it's very sad that people recommend this instead of asking for built-in version, makes me sometimes reconsider my choice to use the language

if they expect language improvements to be templates, what atrocity will be added next?

this is on the same level as std::bool from C++

> The good news is, I have not put a lot of effort so far into micro-optimizing the compile-time performance of match, so there is almost certainly room for improvement.

it is shame that it ended up in the standard library

once Walter will be gone, i will have no faith in the language anymore

All operations that are fed NaN as an operand produce a NaN result. This makes it very obvious when the initialization of a float has been neglected. Defaulting to 0.0 means uninitialization bugs are nearly impossible to detect.

> char default is 0xFF....

Same thing. It's intended to flush out uninitialization bugs.