Re: .ONESHELL enhancement?

[Eric Melski]

David Boyce wrote:
> On Wed, Sep 23, 2009 at 1:43 AM, Eric Melski <address@hidden> wrote:
> >  If you imagine that
> > the command recipe is written to a temporary file and invoked as, eg, "sh
> > script.sh" when .ONESHELL is enabled, it would allow you to use arbitrary
> > interpreters as your shell.
>
> True - but are you imagining or suggesting? I don't think it's likely
> that make will be changed to _actually_ write its recipes to a temp
> file (though what do I know, I'm just an occasional lurker). And if
> you don't use a temp file you have the command line problem (sh -c vs
> perl -e vs ???).

I'm suggesting that this is how .ONESHELL ought to be implemented.  I 
don't see much benefit from your alternative proposal except for your 
build auditor use case, but if we could use an arbitrary interpreter for 
command recipes -- wow!  The possibilities are very exciting, imho.  A 
feature like this might have obviated the creation of tools like Scons, 
for example.

> > This wouldn't work with the "concatenate with &&" proposal.
>
> Yes, but the "concatenate with &&"  proposal seems to have devolved
> into the "don't bother splitting on newlines" proposal anyway, so
> these unfortunate assumptions about the value of $(SHELL) go away.

True.  Note though that writing the recipe to a temp file and invoking 
it with "$(SHELL) tmpfile" would satisfy both our goals, and I think the 
 worry about creating a temp file for each target is overblown.

Cheers,

Eric Melski
http://blog.electric-cloud.com/