On Tue, May 29, 2018 at 12:38:14PM +0200, Thomas Monjalon wrote:
> 28/05/2018 15:20, Bruce Richardson:
> > On Mon, May 28, 2018 at 12:26:24PM +0200, Thomas Monjalon wrote:
> > > 28/05/2018 11:33, Bruce Richardson:
> > > > On Sat, May 26, 2018 at 11:32:53AM +0200, Thomas Monjalon wrote:
> > > > > 25/05/2018 17:18, Bruce Richardson:
> > > > > > On Fri, May 25, 2018 at 04:51:58PM +0200, Thomas Monjalon wrote:
> > > > > > > +load_config () +{ + reset_env + . $(dirname $(readlink
> > > > > > > -e
> > > > > > > $0))/load-devel-config + MESON=${MESON:-meson} +}
> > > > > > Why does this need to be done each time?
> > > > >
> > > > > Because the config could be different for each build (see above).
> > > > >
> > > > How would it be different, it's the same command called with the same
> > > > environment each time?
> > >
> > > No, the idea is to adapt the environment to the build target. As an
> > > example, the dependencies can be different for 32-bit and 64-bit.
> > >
> > I would hope that dependency detection should solve that, but since you
> > already have support for that in existing build script via environment
> > vars, I have no objection to leveraging that in the meson scripts. Overall,
> > though, I'd prefer to ensure that the detection works so that everyone only
> > needs one environment setup in order to get all builds working
> > simultaneously.
>
> The dependency detection cannot work if I have dependencies in uncommon
> directories.
> I think it is important to allow testing compilation with dependencies
> which are available but not installed, by providing paths.
>
Yes, but that should be done via meson configure options to specify the
paths, rather than via environmental variables. AFAIK there is no support in
meson for querying the environment*, every variable that could affect the
build should be explicitly specified
/Bruce
* Ok, no direct way. There are probably ways in which it could be done,
but it's not the meson way to do things.
