Re: [patches] autoconf-it and timestamps

["Peter Seebach" <peter.seebach@xxxxxxxxxxxxx>]

On Wed, 19 Aug 2009 19:07:41 -0500, Joseph S. Myers  
<joseph@xxxxxxxxxxxxxxxx> wrote:

> On Wed, 19 Aug 2009, Peter Seebach wrote:
>
> > My thoughts are roughly that the autoconf-it rule probably shouldn't be
> > modifying the source tree if you're building in another directory --  
> > that just
> > seems dangerous. For our immediate purposes, we're just unilaterally  
> > touching
> > */configure, but that's... uhm. Not good.
>
> Touching the configure scripts and any other checked in generated files  
> is
> what I'd consider the right approach; you want the makefiles to know the
> right commands for regeneration if you did modify configure.in, without
> needing to configure in the source directory.
>
> Perhaps we should add a script like GCC's contrib/gcc_update to handle
> touching generated files.

It turns out to be worse than I thought!  If we touch those files, the  
existing configure scripts produce slightly different output, in  
particular, the libc.info files go in the wrong place.

So what we really want to do is probably rerun autoconf on those in  
advance, once, before doing our multiple builds.  But I sorta wonder if  
this isn't a case where something like gcc's magic for touching copies in  
build directories instead of altering the source tree wouldn't be better.

-s
--
Listen, get this.  Nobody, with a good compiler, needs to be justified.