Implementing a Programming Language in D: Lexical Analysis

[burjui via Digitalmars-d-announce]

On Tuesday, 29 December 2015 at 05:57:34 UTC, Ali Çehreli wrote:
> I've realized that with a nested anonymous enum, there is no 
> need to (and no way of) mentioning the enum type inside a 
> user-defined type. This can simplify the implementation:

Only if you intend to use enum members as manifest constants,
otherwise you're losing the type safety and explicitness
offered by named enums.
For example, I wouldn't use an anonymous enum for lexeme types.

> enum {
>     a,
>     ulong b = 42, c,   // b and c are ulong
>     s = "hello",       // an inferred string between integrals!
>     x = 7, y, z
> }

For me, it's a clear example of unnecessary over-engineering.

> However, move 's' to the end of the list or remove it 
> altogether,
> then x, y, and z become ulong! Weird.

It's even worse than that: x, y and z will still be int (inferred 
from 7).
This code

> void main()
> {
>     enum {
>        a,
>        ulong b = 42, c,   // b and c are ulong
>        x = 7, y, z
>     }
>     import std.stdio, std.typecons;
>     tuple!(typeof(b), typeof(c), typeof(x)).writeln;
> }

will print

> Tuple!(ulong, ulong, int)(0, 0, 0)

which is somewhat counter-intuitive. I would suggest to remove 
this feature
as useless and bug-prone. The following is much clearer IMHO,
even though it's more verbose:

> enum { a }
> enum: ulong { b = 42, c }
> enum { s = "hello" }
> enum { x = 7, y, z }

Even more than that, I would also suggest to remove anonymous 
auto-typed enums
without an initial value from which type can be inferred, e.g.:

> enum { a } // a is an int implicitly

Again, the following is not much harder to write, but the type of 
'a' is immediately clear:

> enum: int { a } // a = int.init, obviously

Although I understand that these are breaking changes,
and D is too mature to break anything (almost not being 
sarcastic).