Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

[Dmitry Gutov]

Hi Juanma,

On 10/22/2015 11:05 AM, Juanma Barranquero wrote:

>  > branch: master
>  > commit 5e4395544c184d9a8702cf05c7aee9329b0b9482
>  > Author: Dmitry Gutov <dgutov@yandex.ru <mailto:dgutov@yandex.ru>>
>  > Commit: Dmitry Gutov <dgutov@yandex.ru <mailto:dgutov@yandex.ru>>
>  >
>  >     Don't declare vc-exec-after anymore
...
> After this change, compiling vc-mtn.el, vc-git.el and vc-bzr.el give
> warnings (the other three don't):

Thank you for pointing that out. Any idea why that happens?

> trunk@C:\...\vc> for %i in (svn mtn hg cvs git bzr) do emacs -batch -Q
> -f batch-byte-compile vc-%i.el
>
> In end of data:
> vc-mtn.el:378:1:Warning: the function `vc-exec-after' might not be
> defined at
>      runtime.

No direct calls to vc-exec-after appear in any of these files anymore. 
All of them call vc-run-delayed, yet, as you say, the warning is 
displayed only when compiling some of them.

I'd think that if the compiler has gotten around to expanding 
vc-run-delayed, it must have already loaded vc-dispatcher. And then it 
also knows there vc-exec-after is defined.

Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

[Juanma Barranquero]

[-- Attachment #1: Type: text/plain, Size: 395 bytes --]

On Thu, Oct 22, 2015 at 2:08 PM, Dmitry Gutov <dgutov@yandex.ru> wrote:

> Any idea why that happens?

Not really. I took a brief look and was puzzled, just like you ;-)

> I'd think that if the compiler has gotten around to expanding
vc-run-delayed, it must have already loaded vc-dispatcher.

Perhaps something is messing with loading vc-dispatcher, though I can't
imagine what or how.

    J

[-- Attachment #2: Type: text/html, Size: 596 bytes --]

Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

[Eli Zaretskii]

> From: Dmitry Gutov <dgutov@yandex.ru>
> Date: Thu, 22 Oct 2015 15:08:37 +0300
> Cc: emacs-devel <emacs-devel@gnu.org>
> 
>     address@hidden:\...\vc> for %i in (svn mtn hg cvs git bzr) do emacs -batch -Q
>     -f batch-byte-compile vc-%i.el
> 
>     In end of data:
>     vc-mtn.el:378:1:Warning: the function `vc-exec-after' might not be
>     defined at
>          runtime.
> 
> No direct calls to vc-exec-after appear in any of these files anymore. All of them call vc-run-delayed, yet, as you say, the warning is displayed only when compiling some of them.

FWIW, on my system I see that warning with all the vc-*.el files that
call vc-run-delayed.  The only exception is vc-svn.el, and the reason
is this single line:

  (autoload 'vc-expand-dirs "vc")

If you remove it, the warning will appear in vc-svn.el as well.
(Loading vc loads vc-dispatcher as a side effect.)

Another data point: replace this:

  (eval-when-compile (require 'vc))

with this:

  (require 'vc)

and the warning also goes way.  So evidently there's some difference
between these two, or maybe a byte-compiler bug.

Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

[Eli Zaretskii]

> Date: Thu, 22 Oct 2015 18:04:39 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: lekktu@gmail.com, emacs-devel@gnu.org
> 
> FWIW, on my system I see that warning with all the vc-*.el files that
> call vc-run-delayed.  The only exception is vc-svn.el, and the reason
> is this single line:
> 
>   (autoload 'vc-expand-dirs "vc")
> 
> If you remove it, the warning will appear in vc-svn.el as well.
> (Loading vc loads vc-dispatcher as a side effect.)
> 
> Another data point: replace this:
> 
>   (eval-when-compile (require 'vc))
> 
> with this:
> 
>   (require 'vc)
> 
> and the warning also goes way.  So evidently there's some difference
> between these two, or maybe a byte-compiler bug.

I think bug#21724 supplies an answer: eval-when-compile makes this
function known _at_compile_time_, but the warning says it might not be
known _at_run_time_.

IOW, for a macro it is enough to use eval-when-compile, but if that
macro calls a function, the function should be either auto-loaded, or
the package that defines that function should be require'd.

Does this make sense?

Re: [Emacs-diffs] master 5e43955 1/2: Don't declare vc-exec-after anymore

[Dmitry Gutov]

On 10/23/2015 10:29 AM, Eli Zaretskii wrote:

> I think bug#21724 supplies an answer: eval-when-compile makes this
> function known _at_compile_time_, but the warning says it might not be
> known _at_run_time_.
>
> IOW, for a macro it is enough to use eval-when-compile, but if that
> macro calls a function, the function should be either auto-loaded, or
> the package that defines that function should be require'd.

That makes a lot of sense, thank you. Even thought it's 
counter-intuitive, because that means we not only have to care about the 
functions we call, we must also know, in this kind of situations, what 
macros expand to.

But the fact that using (autoload foo "pkg") fixes it if pkg loads 
vc-dispatcher, muddies the problem. For me, it makes compilation succeed 
without warnings not only in vc-svn, but in vc-hg and vc-cvs as well.

Unless anyone has a better suggestion (Stefan?), guess I'll just revert 
the commit.