bug#15389: 24.2.91; order of eval-after-load actions

bug#15389: 24.2.91; order of eval-after-load actions

[João Távora]

Hi maintainers,

Consider a file foo-test.el consisting of this content

    (eval-after-load 'foo
      `(progn
         (message "foo1")))
     
    (eval-after-load 'foo
      `(progn
         (message "foo2")))
     
    (provide 'foo) 

If I interactively eval these forms in order using `eval-last-sexp', for
example I get

    foo1
    foo2

in the *Messages* buffer. But if i load the file like so

    emacs -Q --batch -l foo-test.el

I get

    foo2
    foo1

Is this the expected behaviour? Shouldn't the order in which the hooks
are run match the order of definition. 



In GNU Emacs 24.2.91.1 (x86_64-apple-darwin11.4.2, Carbon Version 1.6.0 AppKit 1138.51)
 of 2012-12-25 on king
Windowing system distributor `Apple Inc.', version 10.7.5
Configured using:
 `configure '--prefix=/usr/local/Cellar/emacs-mac/HEAD'
 '--enable-locallisppath=/usr/local/share/emacs/site-lisp'
 '--infodir=/usr/local/Cellar/emacs-mac/HEAD/share/info/emacs'
 '--with-mac' '--enable-mac-app' 'CC=cc''

Important settings:
  value of $LANG: en_US.UTF-8
  locale-coding-system: utf-8-unix
  default enable-multibyte-characters: t

bug#15389: 24.2.91; order of eval-after-load actions

[Glenn Morris]

João Távora wrote:

> Is this the expected behaviour? Shouldn't the order in which the hooks
> are run match the order of definition. 

I don't think eval-after-load promises anything about the order of
evaluation. At the moment, it works like add-hook, ie adds to the front
of the list. Do you have an example where it actually matters?

> In GNU Emacs 24.2.91.1 (x86_64-apple-darwin11.4.2, Carbon Version 1.6.0 AppKit 1138.51)

BTW, there's zero reason to use that anymore since 24.3 was released.

bug#15389: 24.2.91; order of eval-after-load actions

[João Távora]

Glenn Morris <rgm@gnu.org> writes:

> João Távora wrote:
>
>> Is this the expected behaviour? Shouldn't the order in which the hooks
>> are run match the order of definition.
>
> I don't think eval-after-load promises anything about the order of
> evaluation. At the moment, it works like add-hook, ie adds to the front
> of the list.

Well, add-hook has an `append' option right? That would be nice.

> Do you have an example where it actually matters?

Yes. I use small configuration files that are loaded in a particular
order to configure emacs in different flavors, an idea that I took from
debian, i believe, some years ago.

For example, take gnus. If I activate the "common" and "work" flavors in
that order, (which I do when I'm at work) the statements in the "work"
flavor will sometimes overwrite settings that I set in the "common"
flavor, like a different smtp servers.

It all worked fine until I remembered to introduce eval-after-load to
make startup faster.

I could possibly, for instance, never `setq' a variable directly in the
progn body passed to eval-after-load.

Using a million mode-specific hooks I could ensure that "work" always
takes precedence. But this would defeat the purpose of my simple
directories-flavors and 33some-foo-config.el scheme, which has served me
wonderfully for years now and relies on this simple feature of lisp.

But it would just be neater if the order of appearance matched the order
of evaluation.

I don't have to use eval-after-load. Isn't there another similar
mechanism? Where can I start looking?

>> In GNU Emacs 24.2.91.1 (x86_64-apple-darwin11.4.2, Carbon Version 1.6.0 AppKit 1138.51)
>
> BTW, there's zero reason to use that anymore since 24.3 was released.

There's a reason: this one's working fine. It takes me time and patience
to compile another emacs with my fullscreen patches.

Thanks,
João

bug#15389: 24.2.91; order of eval-after-load actions

[Stefan Monnier]

> If I interactively eval these forms in order using `eval-last-sexp', for
> example I get

>     foo1
>     foo2

> in the *Messages* buffer. But if i load the file like so

>     emacs -Q --batch -l foo-test.el

> I get

>     foo2
>     foo1

> Is this the expected behaviour?

No, indeed, thank you for the clear test case.
Should now be fixed in the emacs-24 branch with the trivial patch below.


        Stefan


--- lisp/subr.el	2013-07-03 03:13:07 +0000
+++ lisp/subr.el	2013-09-16 16:28:00 +0000
@@ -1872,7 +1872,7 @@
 				      nil
 				    (remove-hook 'after-load-functions ',fun)
 				    ,',form)))
-		     (add-hook 'after-load-functions fun))
+		     (add-hook 'after-load-functions fun 'append))
 		 ;; Not being provided from a file, run form right now.
 		 ,form)))
       ;; Add FORM to the element unless it's already there.