bug#20139: 25.0.50; eval-buffer ignores lexical-binding

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Jorgen Schaefer]

Given a buffer with these contents:

(setq lexical-binding t)
(message "%s" (lambda () a))

Evaluating both lines with C-x C-e will message (closure (t) nil a), as
it should. But M-x eval-buffer will message (lambda nil a), i.e.
eval-buffer ignores the value of lexical-binding.

I do not think this is the intended behavior :-)

Regards,
Jorgen

In GNU Emacs 25.0.50.1 (x86_64-unknown-linux-gnu)
 of 2015-02-21 on loki.jorgenschaefer.de
Repository revision: 9074a684990600abd9dfad0477c7cd1d2f339ed3
System Description:	Debian GNU/Linux 7.8 (wheezy)

Configured using:
 `configure --without-x'

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Stefan Monnier]

> (setq lexical-binding t)
> (message "%s" (lambda () a))

This is wrong.  The `setq' above is an expression evaluated at run-time,
whereas what you want to say (that this code is written in the lexical
version of the Elisp language) is something that needs to be known before
evaluation proceeds.

You can do M-: (setq lexical-binding t) from that buffer in order to
tell Emacs that the code in the buffer uses the lexical
version of the Elisp language, or better, you should put a "-*-
lexical-binding:t -*-" at the top of the file.


        Stefan

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Jorgen Schäfer]

On Thu, Mar 19, 2015 at 4:10 AM, Stefan Monnier
<monnier@iro.umontreal.ca> wrote:
>> (setq lexical-binding t)
>> (message "%s" (lambda () a))
>
> This is wrong.  The `setq' above is an expression evaluated at run-time,
> whereas what you want to say (that this code is written in the lexical
> version of the Elisp language) is something that needs to be known before
> evaluation proceeds.

I am sorry for the confusion; I added the `setq' form so it is easier
to C-x C-e it. The
bug I describe happens in a buffer with `lexical-binding' set, no
matter which way it
is set. I.e. the following snippet will message (lambda nil t) and
then (closure (t) nil t):

(with-temp-buffer
  (setq lexical-binding t)
  (insert "(message \"%S\" (lambda () t))")
  (eval-buffer)
  (eval-region (point-min) (point-max)))

Regards,
Jorgen

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Nicolas Richard]

Jorgen Schäfer <contact@jorgenschaefer.de> writes:
> . I.e. the following snippet will message (lambda nil t) and
> then (closure (t) nil t):
>
> (with-temp-buffer
>   (setq lexical-binding t)
>   (insert "(message \"%S\" (lambda () t))")
>   (eval-buffer)
>   (eval-region (point-min) (point-max)))

And this will reverse the behaviour :
(with-temp-buffer
  (insert ";; -*- lexical-binding: t; -*-\n(message \"%S\" (lambda () t))")
  (eval-buffer)
  (eval-region (point-min) (point-max)))

IOW eval-buffer obeys the "file"-local variable by calling
    specbind (Qlexical_binding, lisp_file_lexically_bound_p (buf) ? Qt : Qnil);
while eval-region simply uses the current value of lexical-binding.

-- 
Nicolas

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Stefan Monnier]

> IOW eval-buffer obeys the "file"-local variable by calling
>     specbind (Qlexical_binding, lisp_file_lexically_bound_p (buf) ? Qt : Qnil);
> while eval-region simply uses the current value of lexical-binding.

Indeed, and I don't think we can easily change either of these.
Jorgen, what is the original problem you're trying to solve?


        Stefan

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Jorgen Schäfer]

On Thu, Mar 19, 2015 at 3:40 PM, Stefan Monnier
<monnier@iro.umontreal.ca> wrote:
>> I am extracting Emacs Lisp code from a markdown file and evaluating
>> it. I can easily use eval-region instead of eval-buffer, so I have no
>> unsolved problem at the moment. Just figured I'd report this issue.
>
> The basic rule for me is "never (setq lexical-binding ...)".  Instead,
> either add the file-local magic marker and manipulate it as a whole
> buffer/file, or let-bind lexical-binding around the call to
> `eval-region', or pass the second arg to `eval'.

Let-binding `lexical-binding' does not affect `eval-buffer', either.

(with-temp-buffer
  (let ((lexical-binding t))
    (insert "(message \"%S\" (lambda () t))")
    (eval-buffer)
    (eval-region (point-min) (point-max))))

(lambda nil t)
(closure (t) nil t)

Regards,
Jorgen

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Glenn Morris]

Stefan Monnier wrote:

> You can do M-: (setq lexical-binding t) from that buffer in order to
> tell Emacs that the code in the buffer uses the lexical

Apparently that doesn't work: http://debbugs.gnu.org/15070

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Stefan Monnier]

>> The basic rule for me is "never (setq lexical-binding ...)".  Instead,
>> either add the file-local magic marker and manipulate it as a whole
>> buffer/file, or let-bind lexical-binding around the call to
>> `eval-region', or pass the second arg to `eval'.
> Let-binding `lexical-binding' does not affect `eval-buffer', either.

Yes: note that I distinguish between "whole buffer/file" and the other
case (the other case being when you evaluate some chunk of code).


        Stefan

bug#20139: 25.0.50; eval-buffer ignores lexical-binding

[Lars Ingebrigtsen]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

>> IOW eval-buffer obeys the "file"-local variable by calling
>>     specbind (Qlexical_binding, lisp_file_lexically_bound_p (buf) ?
>> Qt : Qnil);
>> while eval-region simply uses the current value of lexical-binding.
>
> Indeed, and I don't think we can easily change either of these.
> Jorgen, what is the original problem you're trying to solve?

So `eval-buffer' ignores the buffer-local value of `lexical-binding',
and that's a feature.  So I think we should just document that?

I've now done so in Emacs 27.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no