user-init-file source vs. compiled

user-init-file source vs. compiled

[Juanma Barranquero]

>From the docstring of `user-init-file':

  File name, including directory, of user's initialization file.
  If the file loaded had extension `.elc', and the corresponding source file
  exists, this variable contains the name of source file, suitable for use
  by functions like `custom-save-all' which edit the init file.

Which is true, *once* the user init file is loaded. While it loads,
`user-init-file' contains the real name of the file loaded:

------ .emacs ------
(message "user-init-file: %S" user-init-file)
--------------------

After compiling .emacs and starting Emacs, *messages* contains:

(C:\emacs\HEAD\bin\emacs.exe)
Loading encoded-kb...done
user-init-file: "c:/home/.emacs.elc"

I've been struggling to find a clear wording to explain it. Suggestions welcome.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Lennart Borgman (gmail)]

Juanma Barranquero wrote:
>> From the docstring of `user-init-file':
> 
>  File name, including directory, of user's initialization file.
>  If the file loaded had extension `.elc', and the corresponding source file
>  exists, this variable contains the name of source file, suitable for use
>  by functions like `custom-save-all' which edit the init file.
> 
> Which is true, *once* the user init file is loaded. While it loads,
> `user-init-file' contains the real name of the file loaded:
> 
> ------ .emacs ------
> (message "user-init-file: %S" user-init-file)
> --------------------
> 
> After compiling .emacs and starting Emacs, *messages* contains:
> 
> (C:\emacs\HEAD\bin\emacs.exe)
> Loading encoded-kb...done
> user-init-file: "c:/home/.emacs.elc"
> 
> I've been struggling to find a clear wording to explain it. Suggestions 
> welcome.


What about the other way round? Changing the behaviour to what the 
documentation says? Is there any reason that user-init-file should not 
always point to the source file?

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:

> What about the other way round? Changing the behaviour to what the
> documentation says? Is there any reason that user-init-file should not
> always point to the source file?

I don't think `load' should worry itself on whether .emacs.el, .emacs,
_emacs.el, _emacs or .emacs.d/init.el exist.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Lennart Borgman (gmail)]

Juanma Barranquero wrote:
> On 2/1/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:
> 
>> What about the other way round? Changing the behaviour to what the
>> documentation says? Is there any reason that user-init-file should not
>> always point to the source file?
> 
> I don't think `load' should worry itself on whether .emacs.el, .emacs,
> _emacs.el, _emacs or .emacs.d/init.el exist.


Could load-file-name be a way for a user to check in user-init-file what 
is actually loaded? Or is that not set here?

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:

> Could load-file-name be a way for a user to check in user-init-file what
> is actually loaded?

What do you mean? `user-init-file' is what is actually loaded.

> Or is that not set here?

Where? Certainly `load-file-name' is set while a file is being loaded.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Lennart Borgman (gmail)]

Juanma Barranquero wrote:
> On 2/1/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:
> 
>> Could load-file-name be a way for a user to check in user-init-file what
>> is actually loaded?
> 
> What do you mean? `user-init-file' is what is actually loaded.
> 
>> Or is that not set here?
> 
> Where? Certainly `load-file-name' is set while a file is being loaded.


I just meant that if the user want to know if he has compiled his user 
init file he/she can look at load-file-name. There is no need to give 
this information also through user-init-file during the load of 
user-init-file. So user-init-file could from that point of view point to 
the source file.

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:

> So user-init-file could from that point of view point to
> the source file.

At the risk of repeating myself: I don't think `load' should worry
with determining what is the right init source file.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Stefan Monnier]

>  File name, including directory, of user's initialization file.
>  If the file loaded had extension `.elc', and the corresponding source file
>  exists, this variable contains the name of source file, suitable for use
>  by functions like `custom-save-all' which edit the init file.

I think we should *strongly* recommend not to compile the .emacs file.
It's a dumb thing to do, unless you're a masochist anyway.


        Stefan

Re: user-init-file source vs. compiled

[Chong Yidong]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>>  File name, including directory, of user's initialization file.
>>  If the file loaded had extension `.elc', and the corresponding source file
>>  exists, this variable contains the name of source file, suitable for use
>>  by functions like `custom-save-all' which edit the init file.
>
> I think we should *strongly* recommend not to compile the .emacs file.
> It's a dumb thing to do, unless you're a masochist anyway.

Currently, the Emacs manual says

     If you have a large amount of code in your `.emacs' file, you
  should rename it to `~/.emacs.el', and byte-compile it.  *Note Byte
  Compilation: (elisp)Byte Compilation, for more information about
  compiling Emacs Lisp programs.

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

> I think we should *strongly* recommend not to compile the .emacs file.

Perhaps. But Emacs supports having .emacs compiled anyway, so the
docstring for `user-init-file' should be accurate.

> It's a dumb thing to do, unless you're a masochist anyway.

Why? I have a medium-sized .emacs (~1800 lines) and it loads faster
when compiled (or at least that's the feeling I get).

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Eli Zaretskii]

> Date: Thu, 1 Feb 2007 16:10:37 +0100
> From: "Juanma Barranquero" <lekktu@gmail.com>
> 
>   File name, including directory, of user's initialization file.
>   If the file loaded had extension `.elc', and the corresponding source file
>   exists, this variable contains the name of source file, suitable for use
>   by functions like `custom-save-all' which edit the init file.
> 
> Which is true, *once* the user init file is loaded. While it loads,
> `user-init-file' contains the real name of the file loaded:
> 
> ------ .emacs ------
> (message "user-init-file: %S" user-init-file)
> --------------------
> 
> After compiling .emacs and starting Emacs, *messages* contains:
> 
> (C:\emacs\HEAD\bin\emacs.exe)
> Loading encoded-kb...done
> user-init-file: "c:/home/.emacs.elc"
> 
> I've been struggling to find a clear wording to explain it. Suggestions welcome.

How about adding this to the doc string:

  While Emacs loads and evaluates the init file, value is the real
  name of the file, regardless of whether or not it has the `.elc'
  extension.

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Eli Zaretskii <eliz@gnu.org> wrote:

> How about adding this to the doc string:

Thanks!

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Stefan Monnier]

>>> File name, including directory, of user's initialization file.
>>> If the file loaded had extension `.elc', and the corresponding source file
>>> exists, this variable contains the name of source file, suitable for use
>>> by functions like `custom-save-all' which edit the init file.
>> 
>> I think we should *strongly* recommend not to compile the .emacs file.
>> It's a dumb thing to do, unless you're a masochist anyway.

> Currently, the Emacs manual says

>      If you have a large amount of code in your `.emacs' file, you
>   should rename it to `~/.emacs.el', and byte-compile it.  *Note Byte
>   Compilation: (elisp)Byte Compilation, for more information about
>   compiling Emacs Lisp programs.

I clearly disagree.  If there's no loop in your .emacs, then byte-compiling
won't bring you any benefit.  And if there are loops and byte-compiling is
important to make them fast enough, then you should most likely take those
things out of your .emacs and move them to a separate file/package.


        Stefan

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/1/07, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

> I clearly disagree.  If there's no loop in your .emacs, then byte-compiling
> won't bring you any benefit.

It's not about the code being faster, is about the .elc being smaller
(about 10% smaller) than the .el.

> then you should most likely take those
> things out of your .emacs and move them to a separate file/package.

Why? You prefer it that way, OK, do that. I don't like having lots of
small .el files loaded from .emacs; it's easier to maintain it
up-to-date between two or more computers when it's a single file.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Stefan Monnier]

>> I clearly disagree.  If there's no loop in your .emacs, then byte-compiling
>> won't bring you any benefit.
> It's not about the code being faster, is about the .elc being smaller
> (about 10% smaller) than the .el.

10% smaller will result in a non-measurably faster load, and will still take
90% more disk space.  So what's the advantage again?

>> then you should most likely take those things out of your .emacs and move
>> them to a separate file/package.

> Why? You prefer it that way, OK, do that. I don't like having lots of
> small .el files loaded from .emacs; it's easier to maintain it
> up-to-date between two or more computers when it's a single file.

I don't want to make it impossible to byte-compile your .emacs.  If that's
really what you want, go for it.  I'm just saying that we should discourage
people from doing that, because too many users think "oh Emacs takes a long
time to start, I bet if I byte-compile it it'll start faster", but it turns
out it's not faster and next time they forget to re-compile their .emacs
they get bitten.  It's not the most common problem on gnu.emacs.help, but
it's sufficiently common that I think the potential very minor advantage(s)
is just not worth it.

OTOH I'm in favor of adding some kind of hack so that the .emacs file
automatically goes through the byte-compiler's sanity checks when you save
it (so they user might can warnings about using obsolete
variables/functions).


        Stefan

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/2/07, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

> 10% smaller will result in a non-measurably faster load, and will still take
> 90% more disk space.  So what's the advantage again?

The wonderful, perhaps subjective feeling that it is loading faster :)

> I don't want to make it impossible to byte-compile your .emacs.  If that's
> really what you want, go for it.

I've been using Emacs for nine years and I just started to
byte-compile my .emacs a week ago, so it's not a matter of me "really
wanting". This thread was about a variable's description being
inaccurate (fortunately it is now fixed).

> I'm just saying that we should discourage
> people from doing that, because too many users think "oh Emacs takes a long
> time to start, I bet if I byte-compile it it'll start faster", but it turns
> out it's not faster and next time they forget to re-compile their .emacs
> they get bitten.

I've added this to my site-start.el:

(catch 'init-file
  (dolist (source-file '("~/.emacs.el" "~/.emacs." "~/_emacs.el"
                         "~/_emacs" "~/.emacs.d/init.el"))
    (when (file-exists-p source-file)
      (require 'bytecomp)
      (unless (fboundp 'time-less-p) (require 'time-date))
      (let ((byte-file (byte-compile-dest-file source-file)))
        (when (and (file-exists-p byte-file)
                   (time-less-p (nth 5 (file-attributes byte-file))
                                (nth 5 (file-attributes source-file))))
          (unless (byte-compile-file source-file)
            (setq init-file-user nil))))
      (throw 'init-file t))))

which recompiles my .emacs.el if needed (the compilation log is quite
visible, so I don't miss that something happened), with the added
benefit that .emacs(.elc?)? is not loaded if the compilation fails.

> It's not the most common problem on gnu.emacs.help, but
> it's sufficiently common that I think the potential very minor advantage(s)
> is just not worth it.

By all means, if you think users should not compile their .emacs
unless they know what they're doing, just add a prominent notice to
the relevant docs. You're talking about what should we recommend; I
was just talking about not misleading people, given the fact that
bytecompiling .emacs is *allowed*. I have no opinion on whether we
should en- or discourage bytecompiling the init file.

> OTOH I'm in favor of adding some kind of hack so that the .emacs file
> automatically goes through the byte-compiler's sanity checks when you save
> it (so they user might can warnings about using obsolete
> variables/functions).

That's a really nice idea.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Lennart Borgman (gmail)]

Juanma Barranquero wrote:

>> OTOH I'm in favor of adding some kind of hack so that the .emacs file
>> automatically goes through the byte-compiler's sanity checks when you 
>> save
>> it (so they user might can warnings about using obsolete
>> variables/functions).
> 
> That's a really nice idea.


But is not a lot more elisp modules loaded at that time? (Though perhaps 
you can workaround that too.)

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/2/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:

> But is not a lot more elisp modules loaded at that time?

A few modules, yes, but do you really save .emacs that much?

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Stefan Monnier]

> This thread was about a variable's description being
> inaccurate (fortunately it is now fixed).

Yes, and I'm glad it's been fixed.  I just hijacked the thread for some
other (related) purpose.
We don't seem to be in disagreement.


        Stefan

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/2/07, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

> We don't seem to be in disagreement.

I don't think so, no.

                    /L/e/k/t/u

Re: user-init-file source vs. compiled

[Lennart Borgman (gmail)]

Juanma Barranquero wrote:
> On 2/2/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:
> 
>> But is not a lot more elisp modules loaded at that time?
> 
> A few modules, yes, but do you really save .emacs that much?


Eh, no, but I have usually loaded it before I save it ... ;-)

Re: user-init-file source vs. compiled

[Juanma Barranquero]

On 2/2/07, Lennart Borgman (gmail) <lennart.borgman@gmail.com> wrote:

> Eh, no, but I have usually loaded it before I save it ... ;-)

Hmm. I don't understand. Stefan proposed a mechanism to check .emacs
when it is written. What's that got to do with when/how you load it?

                    /L/e/k/t/u