Almost keywords

[Duncan Booth]

bearophileHUGS at lycos.com wrote:

> A possible way to avoid such bugs is to turn all those names like
> "list", "map", "filter", "self", etc into keywords. But this may have
> some disadvantages (well, I think I'd like to have "self" as keyword,
> seen how all Python books strong suggest to not use a name different
> from "self").

I'm not convinced by making 'self' a keyword. Forcing the first argument of 
a method to be named 'self' fails for many reasons (e.g. 
classmethod/staticmethod) so I don't really see a benefit to making it 
special in any way.

For the other names though, this would have the advantage of allowing the 
compiler to optimise access to builtins which should provide a slight speed 
improvement and I think would also make life easier for some 
implementations (especially things like 'locals()').

A half-way house would be to generate a warning for any use of a name which 
masks one of the reserved builtins, but only optimise lookups in scopes 
where no such masking is seen by the compiler: this would prevent you 
overriding builtins by injecting a name into the global namespace from 
outside a module but not otherwise break existing code.

Of course any of this would break some existing code and therefore isn't 
going to happen.