Re: [PATCH] * src/eval.c: Stop checking for nvars, and use only CONSP

[Pip Cet <pipcet@gmail.com>]

On Tue, Mar 2, 2021 at 3:48 PM Stefan Monnier <monnier@iro.umontreal.ca> wrote:
> > I'm all for a revolution, but it might be a bit early to chop off this
> > particular king's head...
> Until we have good debugging support for byte-compiled code, the
> interpreter isn't going anywhere, indeed.

We could rewrite it in ELisp, though :-)

> But error reporting from the interpreter is very secondary because in my
> view that interpreter should only be used for bootstrap and for
> debugging, so all the code it executes should *also* be byte-compiled
> (so we can rely on the byte-compiler for error reporting).

Agreed. But having to choose between compile-time error reporting and
run-time debugging can be very inconvenient, and now we're about to
add the third choice of "debug the natively-compiled code in gdb",
everything's going to get even more confusing...

> >>     (defun eval (exp) (funcall (byte-compile `(lambda () ,exp))))
> > Except for the ones that can't.  ELC is still limited to 64K constants
> > in the vector, for example, isn't it?
>
> I believe so, yes.  If/when we bump into this limit we can push it
> further (or finally replace our bytecode language with a new one ;-).

I hear the hot new thing is to use leb128 integers for everything.

> > But as for the original question, do we have to have Flet?
>
> I have not seen this question asked in this thread.

The question was "can we remove this code from Flet?", right? I think
"yes, all of it. No, all of Flet." is a valid answer to that question
:-)

(I dislike the "let" family, mostly for the fact that it is a family.
It leads naturally to cl-loop.)

Pip