Re: [PATCH] configure: warn if not using a separate build directory

[Daniel P . Berrangé]

On Tue, Mar 31, 2020 at 09:44:37AM -0500, Eric Blake wrote:
> On 3/31/20 5:37 AM, Daniel P. Berrangé wrote:
> > Running configure directly from the source directory is a build
> > configuration that will go away in future. It is also not currently
> > covered by any automated testing. Display a deprecation warning if
> > the user attempts to use an in-srcdir build setup, so that they are
> > aware that they're building QEMU in an undesirable manner.
> > 
> > Signed-off-by: Daniel P. Berrangé <address@hidden>
> > ---
> >   configure | 30 ++++++++++++++++++++++++++++++
> >   1 file changed, 30 insertions(+)
> > 
> 
> It sounds like you already have a v2 coming up to address Phillipe's
> comments, but I'd be happy to see this concept make it into v5.0.
> 
> > diff --git a/configure b/configure
> > index e225a1e3ff..1ab7492ab5 100755
> > --- a/configure
> > +++ b/configure
> > @@ -3,6 +3,19 @@
> >   # qemu configure script (c) 2003 Fabrice Bellard
> >   #
> > +BUILDDIR=$(pwd)
> 
> Why fork out to 'pwd', when we can rely on $PWD for the same answer? Note
> that both $(pwd) and $PWD $BUILDDIR always give an absolute (but no
> necessarily canonical) path, which means...
> 
> > +SRCDIR=$(dirname "$0")
> > +
> > +ABS_BUILDDIR=$(realpath $BUILDDIR)
> 
> ...calling this ABS_BUILDDIR is a misnomer (it was already absolute). What
> it is really doing is chasing through symlinks to result in a canonical
> name, particularly since unless you have absolute names, using only string
> comparison to see if two filenames are equivalent will have false negatives
> (/tmp vs. /tmp/., for example).
> 
> > +ABS_SRCDIR=$(realpath $SRCDIR)
> 
> Not robust if there are spaces in the directory names.  Safer would be
> ABS_BUILDDIR=$(realpath -- "$BUILDDIR")
> and similarly for ABS_SRCDIR.
> 
> > +
> > +in_srcdir=no
> > +if [ "$ABS_SRCDIR" == "$ABS_BUILDDIR" ]
> 
> Bashism. You MUST spell this '=', not '==', since configure is run under
> /bin/sh which might be dash which does not understand ==.

Ok, will address all of the above.

> > @@ -6799,6 +6812,23 @@ if test "$supported_os" = "no"; then
> >       echo "us upstream at address@hidden."
> >   fi
> > +if test "$in_srcdir" = "yes"; then
> > +    echo
> > +    echo "WARNING: SUPPORT FOR IN SOURCE DIR BUILDS IS DEPRECATED"
> > +    echo
> > +    echo "Support for running the 'configure' script directly from the"
> > +    echo "source directory is deprecated and will go away in a future"
> > +    echo "release. In source dir builds are not covered by automated"
> > +    echo "testing and are liable to break without warning. Users are"
> > +    echo "strongly recommended to switch to a separate build directory:"
> > +    echo
> > +    echo "  $ mkdir build"
> > +    echo "  $ cd build"
> > +    echo "  $ ../configure"
> > +    echo "  $ make"
> > +    echo
> > +fi
> 
> You know, it WOULD be possible to further enhance this to actually create
> 'build' as well as a shim GNUmakefile that would auto-forward on to building
> directly in build, so that you maintain the illusion of an in-place build
> (other than all the build artifacts now living in a different location), and
> muscle memory for in-tree 'make' still works; I've posted elsewhere and will
> repeat here the contents of my GNUmakefile:

I wanted to focus strictly on the part that we have agreement on, namely
the deprecation. Any functional changes should be completely separate
to avoid holding up the merge of the warning message patch.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|