Automatic bootstrap

Automatic bootstrap

[Richard M Stallman]

A week or so there was a discussion about the inconvenience of the
existence of `make bootstrap'.  I suggested that we create a source
file bootstrap.el whose compiled file, bootstrap.elc, would control
whether `make emacs' does a bootstrap or not.  We could also change
bootstrap.el trivially after installing any change which we think
makes it advisable for testers to bootstrap afresh.

I hope that this would eliminate the need for even developers to know
about `make boostrap'.

There was no reaction to this.  What do people think of it?

Re: Automatic bootstrap

[Jason Rumney]

Richard M Stallman wrote:
> A week or so there was a discussion about the inconvenience of the
> existence of `make bootstrap'.  I suggested that we create a source
> file bootstrap.el whose compiled file, bootstrap.elc, would control
> whether `make emacs' does a bootstrap or not.  We could also change
> bootstrap.el trivially after installing any change which we think
> makes it advisable for testers to bootstrap afresh.
>   

I don't think it will work, as developers who make changes that make it 
advisable to bootstrap would forget to change that file. Other 
developers might change it needlessly, causing testers with low end 
hardware unnecessary pain.

Re: Automatic bootstrap

[Paul Michael Reilly]

Richard M Stallman wrote:
> A week or so there was a discussion about the inconvenience of the
> existence of `make bootstrap'.  I suggested that we create a source
> file bootstrap.el whose compiled file, bootstrap.elc, would control
> whether `make emacs' does a bootstrap or not.  We could also change
> bootstrap.el trivially after installing any change which we think
> makes it advisable for testers to bootstrap afresh.

What would be the content of bootstrap.el?

-pmr

Re: Automatic bootstrap

[Stefan Monnier]

>> A week or so there was a discussion about the inconvenience of the
>> existence of `make bootstrap'.  I suggested that we create a source
>> file bootstrap.el whose compiled file, bootstrap.elc, would control
>> whether `make emacs' does a bootstrap or not.  We could also change
>> bootstrap.el trivially after installing any change which we think
>> makes it advisable for testers to bootstrap afresh.

> I don't think it will work, as developers who make changes that make it
> advisable to bootstrap would forget to change that file.  Other developers
> might change it needlessly, causing testers with low end hardware
> unnecessary pain.

I tend to agree.  BTW, I'm playing around right now with a slightly
different dependency structure to make "make" do the bootstrap
automatically, so that "make bootstrap" is turned into "make
bootstrap-clean; ./config.status; make".  Basically, we always build
both `emacs' and `bootstrap-emacs', where `bootstrap-emacs' is the
executable used to compile the preloaded files.  The code I have now
kinda works, but it has some downsides.  Also, it's tremendously
difficult to make this work efficiently in the sequential case (avoid
redundant dumps and byte-compilations) and in the parallel case as well
(e.g. avoid building `bootstrap-emacs' multiple times in parallel).


        Stefan

Re: Automatic bootstrap

[Richard M Stallman]

    What would be the content of bootstrap.el?

My proposal does not require it to contain anything,
so it could be empty.

Re: Automatic bootstrap

[Miles Bader]

Richard M Stallman <rms@gnu.org> writes:
>     What would be the content of bootstrap.el?
>
> My proposal does not require it to contain anything,
> so it could be empty.

Why use the extension ".el" then?  It seems misleading.

-Miles

-- 
Un-American, adj. Wicked, intolerable, heathenish.

Re: Automatic bootstrap

[Richard M Stallman]

    > A week or so there was a discussion about the inconvenience of the
    > existence of `make bootstrap'.  I suggested that we create a source
    > file bootstrap.el whose compiled file, bootstrap.elc, would control
    > whether `make emacs' does a bootstrap or not.  We could also change
    > bootstrap.el trivially after installing any change which we think
    > makes it advisable for testers to bootstrap afresh.
    >   

    I don't think it will work, as developers who make changes that make it 
    advisable to bootstrap would forget to change that file.

Indeed, that might happen.  But is that a reason to reject the idea?
I don't think so.  Even if things are not perfect, they will be better.

     Other 
    developers might change it needlessly, causing testers with low end 
    hardware unnecessary pain.

I don't think that is likely to happen very often.
People have no reason to do this.

Re: Automatic bootstrap

[Richard M Stallman]

    > My proposal does not require it to contain anything,
    > so it could be empty.

    Why use the extension ".el" then?  It seems misleading.

Because bootstrapping generates .elc files from all the .el files.
It will do this one just like the rest.
Making this a .el file is therefore natural.

Likewise, a release comes with the .elc files,
so it will come with this one, and that will prevent
bootstrapping when people build the release.