Re: Emacs rewrite in a maintainable language

[Oleh Krehel]

David Kastrup <address@hidden> writes:

> Oleh Krehel <address@hidden> writes:
>> I think slowly moving the Emacs C core to C++ is a good idea.
>
> I don't, and the reasons are similar to why Linux on C++ crashed and
> burned.

I haven't investigated that topic. Could you give the summary of the
reason?

> The theory is very nice, but C++ is not a stable basis.  Every 5 years
> it gets completely redefined, and the Elisp semantics don't magically
> move with that.

Not completely. I'm thinking of using only a very small subset of C++ /
superset of C:

- the C features
- classes with getters/setters (no need for inheritance)
- typed constants
- templated functions where appropriate

Essentially what good C programs do anyway, but with proper support from
the compiler. These features are supported by any version of any C++
compiler.

>> I've been trying to get into understanding Emacs C core, but it's just
>> so hard with all those macros around. I think even replacing `#define`
>> with `const` would be a huge improvement, worthwhile of the switch.
>> The second step would be to replace some macros with type checked
>> template functions.
>
> "template functions"?  Seriously?

Yes, very.

Quoting the Emacs sources:

    # define MIN(a,b) ((a) < (b) ? (a) : (b))
    #define streq(X, Y) (*(X) == *(Y) && strcmp ((X) + 1, (Y) + 1) == 0)
    #define SET_FLAG(F, FLAG)   ((F) |= (FLAG))
    #define HAS_FLAG(F, FLAG)   (((F) & (FLAG)) != 0)
    #define xnew(n, Type)      ((Type *) xmalloc ((n) * sizeof (Type)))
    #define xrnew(op, n, Type) ((op) = (Type *) xrealloc (op, (n) * sizeof 
(Type)))
    #define switch_line_buffers() (curndx = 1 - curndx)
    #define curlinepos (lbs[curndx].linepos)
    #define BVAR(buf, field) ((buf)->field ## _)

Some of these can be replaced with plain type checked C++
functions. Others can be replaced with templated C++ functions, that are
still type checked. Still others with public getters and setters.

    Oleh