Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Juanma Barranquero]

On Thu, Apr 10, 2008 at 10:59 AM, Jason Rumney <jasonr@gnu.org> wrote:

>  +  make recompile
>  +  Recompiles any changed lisp files after a cvs update.  This saves
>  +  doing a full bootstrap after every update.  If this or a subsequent
>  +  make fail, you probably need to perform a full bootstrap, though
>  +  running this target multiple times may eventually sort out the
>  +  interdependencies.

Currently, the `recompile' target of nt/makefile.w32-in runs the
`recompile' target of lisp/makefile. It would be perhaps more useful
to run the `cvs-update' target introduced by Eli:

  # This is useful after "cvs up".
  cvs-update: recompile autoloads finder-data custom-deps

BTW, what is the `updates' target of lisp/Makefile.in (and
makefile.w32-in) used for?

  updates: update-subdirs autoloads finder-data custom-deps

Perhaps `cvs-update' should just be:

  cvs-update: recompile updates

as, after a cvs up, it makes sense to do update-subdirs.

 Juanma

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Jason Rumney]

Juanma Barranquero wrote:
> On Thu, Apr 10, 2008 at 10:59 AM, Jason Rumney <jasonr@gnu.org> wrote:
>
>   
>>  +  make recompile
>>  +  Recompiles any changed lisp files after a cvs update.  This saves
>>  +  doing a full bootstrap after every update.  If this or a subsequent
>>  +  make fail, you probably need to perform a full bootstrap, though
>>  +  running this target multiple times may eventually sort out the
>>  +  interdependencies.
>>     
>
> Currently, the `recompile' target of nt/makefile.w32-in runs the
> `recompile' target of lisp/makefile. It would be perhaps more useful
> to run the `cvs-update' target introduced by Eli:
>   

That is not what INSTALL.CVS recommends. Perhaps it should.

Recommended build procedure [was Re: [Emacs-diffs] Changes to emacs/nt/INSTALL, v]

[Glenn Morris]

Jason Rumney wrote:

>> Currently, the `recompile' target of nt/makefile.w32-in runs the
>> `recompile' target of lisp/makefile. It would be perhaps more useful
>> to run the `cvs-update' target introduced by Eli:
>>   
>
> That is not what INSTALL.CVS recommends. Perhaps it should.

I think INSTALL.CVS should as a default recommend the most thorough
way to rebuild. Then say "if you want to save time, you can try X,Y,Z;
but it may not work".

CPU time is cheap, and the time it takes to fully build Emacs is
nothing special compared to other modern applications. Shortcuts just
confuse people when they go wrong, and lead to lots of emails about
spurious problems. The time it takes to sort out people's confusion is
more expensive.

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Eli Zaretskii]

> Date: Thu, 10 Apr 2008 11:36:58 +0200
> From: "Juanma Barranquero" <lekktu@gmail.com>
> Cc: Eli Zaretskii <eliz@gnu.org>, Emacs Devel <emacs-devel@gnu.org>
> 
>   updates: update-subdirs autoloads finder-data custom-deps
> 
> Perhaps `cvs-update' should just be:
> 
>   cvs-update: recompile updates
> 
> as, after a cvs up, it makes sense to do update-subdirs.

Usually, update-subdirs is just a waste of time, since new directories
are seldom created.

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Jason Rumney]

Eli Zaretskii wrote:
>> Date: Thu, 10 Apr 2008 11:36:58 +0200
>> From: "Juanma Barranquero" <lekktu@gmail.com>
>> Cc: Eli Zaretskii <eliz@gnu.org>, Emacs Devel <emacs-devel@gnu.org>
>>
>>   updates: update-subdirs autoloads finder-data custom-deps
>>
>> Perhaps `cvs-update' should just be:
>>
>>   cvs-update: recompile updates
>>
>> as, after a cvs up, it makes sense to do update-subdirs.
>>     
>
> Usually, update-subdirs is just a waste of time, since new directories
> are seldom created.
>   

But its a waste of a very little amount of time, and a new subdirectory 
was created for org-mode just a couple of weeks ago, so seldom is not never.

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Juanma Barranquero]

On Thu, Apr 10, 2008 at 9:52 PM, Eli Zaretskii <eliz@gnu.org> wrote:

>  Usually, update-subdirs is just a waste of time, since new directories
>  are seldom created.

Well, I'd say the point of cvs-update is to be correct over fast,
isn't it? And as Jason points out, it is a negligible amount of time
anyway.

 Juanma

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Eli Zaretskii]

> Date: Fri, 11 Apr 2008 00:24:51 +0200
> From: "Juanma Barranquero" <lekktu@gmail.com>
> Cc: jasonr@gnu.org, emacs-devel@gnu.org
> 
> On Thu, Apr 10, 2008 at 9:52 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
> >  Usually, update-subdirs is just a waste of time, since new directories
> >  are seldom created.
> 
> Well, I'd say the point of cvs-update is to be correct over fast,
> isn't it?

No, it isn't.  If you want to be correct, use "make bootstrap".
cvs-update was introduced as a very useful shortcut, which takes half
a minute, if only a few Lisp files were modified, rather than one and
a half hour taken by bootstrapping.

> And as Jason points out, it is a negligible amount of time
> anyway.

One (minor) problem is that it regenerates subdirs.el unconditionally,
so whatever targets that depend on it will be remade as well.

But these are all minor considerations, admittedly.

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Juanma Barranquero]

On Fri, Apr 11, 2008 at 10:43 AM, Eli Zaretskii <eliz@gnu.org> wrote:

>  No, it isn't.  If you want to be correct, use "make bootstrap".

I do, daily :) But still I use cvs-update quite often.

>  cvs-update was introduced as a very useful shortcut, which takes half
>  a minute, if only a few Lisp files were modified

Hm. I would've said that the whole point of cvs-update is to be more
correct than recompile; else, why not simply use recompile, which is
faster?

> rather than one and
> a half hour taken by bootstrapping.

Just out of curiosity: in what kind of hardware? On a 3GHz Pentium 4,
it takes less than half an hour for me, from `make maintainer-clean'
to installation (though I cut some time by always doing an in-place
installation). That with MinGW's gcc; when I used Visual C it was even
faster.

>  One (minor) problem is that it regenerates subdirs.el unconditionally,
>  so whatever targets that depend on it will be remade as well.

Aha.

>  But these are all minor considerations, admittedly.

Yes. I'm more interested in the question of whether nt makefile's
recompile should call lisp makefile's cvs-update, or perhaps recompile
+ updates.

 Juanma

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Eli Zaretskii]

> Date: Fri, 11 Apr 2008 11:13:05 +0200
> From: "Juanma Barranquero" <lekktu@gmail.com>
> Cc: jasonr@gnu.org, emacs-devel@gnu.org
> 
> >  cvs-update was introduced as a very useful shortcut, which takes half
> >  a minute, if only a few Lisp files were modified
> 
> Hm. I would've said that the whole point of cvs-update is to be more
> correct than recompile; else, why not simply use recompile, which is
> faster?

Because without the other prerequisites of cvs-update, especially
autoloads, you will have a buggy Emacs.

> > rather than one and
> > a half hour taken by bootstrapping.
> 
> Just out of curiosity: in what kind of hardware? On a 3GHz Pentium 4,
> it takes less than half an hour for me, from `make maintainer-clean'
> to installation (though I cut some time by always doing an in-place
> installation). That with MinGW's gcc; when I used Visual C it was even
> faster.

It's the same hardware, so maybe half an hour is so long for me that I
took it for three times that ;-)  I don't really remember how long it
was exactly, but it seemed forever.

> I'm more interested in the question of whether nt makefile's
> recompile should call lisp makefile's cvs-update, or perhaps recompile
> + updates.

I never do a "make recompile" from the nt subdirectory.

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Juanma Barranquero]

On Fri, Apr 11, 2008 at 12:22 PM, Eli Zaretskii <eliz@gnu.org> wrote:

>  > Hm. I would've said that the whole point of cvs-update is to be more
>  > correct than recompile; else, why not simply use recompile, which is
>  > faster?
>
>  Because without the other prerequisites of cvs-update, especially
>  autoloads, you will have a buggy Emacs.

I thought that was implied in "to be more correct than recompile" ;-)

>  It's the same hardware, so maybe half an hour is so long for me that I
>  took it for three times that ;-)  I don't really remember how long it
>  was exactly, but it seemed forever.

Oh, I usually do (in a .BAT)

   cvs update
   cd nt
   make maintainer-clean  ; realclean before
   configure
   make bootstrap install

as the first thing in the morning, and I let it happily compile. Time
is not much of an issue, I use 22.X meanwhile.

>  I never do a "make recompile" from the nt subdirectory.

I don't, either (I use cvs-update from lisp/), but as of yesterday,
nt/INSTALL includes this:

  make recompile
  Recompiles any changed lisp files after a cvs update.  This saves
  doing a full bootstrap after every update.  If this or a subsequent
  make fail, you probably need to perform a full bootstrap, though
  running this target multiple times may eventually sort out the
  interdependencies.

so it would be perhaps preferable for recompile (from nt/) to do a bit
more work and be more correct.

Just a nitpick, though.

   Juanma

Re: [Emacs-diffs] Changes to emacs/nt/INSTALL,v

[Eli Zaretskii]

> Date: Fri, 11 Apr 2008 12:46:03 +0200
> From: "Juanma Barranquero" <lekktu@gmail.com>
> Cc: jasonr@gnu.org, emacs-devel@gnu.org
> 
> On Fri, Apr 11, 2008 at 12:22 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
> >  > Hm. I would've said that the whole point of cvs-update is to be more
> >  > correct than recompile; else, why not simply use recompile, which is
> >  > faster?
> >
> >  Because without the other prerequisites of cvs-update, especially
> >  autoloads, you will have a buggy Emacs.
> 
> I thought that was implied in "to be more correct than recompile" ;-)

It is more correct, but it does not aim to be absolutely correct.  It
is okay for it to fail in rare situations, if that saves a significant
amount of time.