Re: Bash vs. sh

[Joel E. Denny]

On Wed, 23 Apr 2008, Eric Blake wrote:

> I'm not quite sure what is best to do here.  It seems like the simplest
> wrapper might be this untested script (short enough to possibly be worth
> distributing without copyright notice, since adding the notice would
> double its size):
> 
> #!/bin/sh
> if test "${GNULIB_SRCDIR+set}" != set ; then
> ~  GNULIB_SRCDIR=`sed -n 's/gnulib_dir = //p' < cfg.mk`
> fi
> if autom4te --language=m4sh -o bootstrap_inner \
> ~    "${GNULIB_SRCDIR}/bootstrap.m4sh" ; then
> ~  exec bootstrap_inner
> else
> ~  status=$?
> ~  echo "$0: Please follow the directions in README-hacking to" \
> ~    "bootstrap this package" >&2
> ~  (exit $status); exit $status
> fi

What about something like the following?

  #!/bin/sh
  # If you have trouble using this script, please see README-hacking.
  if test "${GNULIB_SRCDIR+set}" != set ; then
    GNULIB_SRCDIR=`sed -n 's/gnulib_dir = //p' < cfg.mk`
  fi
  if test -d "$GNULIB_SRCDIR"; then
    rm -f bootstrap.m4sh
    ln -s "$GNULIB_SRCDIR/bootstrap.m4sh" bootstrap.m4sh || exit 1
  else
    # git, wget, or cvs command to download bootstrap.m4sh || exit 1
  fi
  autom4te --language=m4sh -o bootstrap-inner bootstrap.m4sh || exit 1
  exec bootstrap-inner

The main difference is that, if you haven't downloaded gnulib yet, you can 
still run bootstrap.  bootstrap will download enough to generate 
bootstrap-inner, which will download the rest of gnulib just as the old 
bootstrap does.

I'm new to git.  I looked through git's help for a while, and I didn't 
find a command to simply export a file from a remote repository.  Maybe I 
missed it.  Regardless, wget can fetch from savannah.  I think that will 
be sufficient.

I also eliminated the README-hacking warning.  I figured its main purpose 
was to help the user who doesn't know about downloading gnulib, but that 
download is handled automatically now.

> This isn't very robust to failure.  But the more robust you make it, the
> more it argues for having a single bootstrap script in the first place.

I agree.  Moreover, if it stays short and conceptually simple (just do 
enough to generate bootstrap-inner and run it), I see no need to add a lot 
of careful diagnostics because project developers can then easily diagnose 
failures themselves.

> Should bootstrap maintain both bootstrap.m4sh and the generated bootstrap?

By eventually creating a symlink to the bootstrap.m4sh within the full 
gnulib download, I think my approach takes care of this.

> ~ Should projects that want to use gnulib's bootstrap also check it into
> their own repository, or just use the README files to inform users that
> their package can't be bootstrapped until gnulib's bootstrap file is run?

I don't see a need for either.  Provided the user has basic tools 
(autoconf, wget, etc.), he merely runs ./bootstrap.

If a project wants to maintain its own version of bootstrap.m4sh and 
periodically perform a manual sync with gnulib's, it can by rewriting its 
outer bootstrap not to use gnulib's.  A better approach would be to extend 
its outer bootstrap to perform whatever additional tasks that specific 
project needs after running gnulib's bootstrap.  Even better would be to 
contribute to gnulib.