Re: [PATCH v3 3/3] build: RFC - add support for optional dependencies

[Stephen Hemminger <stephen@networkplumber.org>]

On Wed, 20 Dec 2023 15:43:36 +0000
Bruce Richardson <bruce.richardson@intel.com> wrote:

> On Wed, Dec 20, 2023 at 04:08:08PM +0100, Morten Brørup wrote:
> > > From: Bruce Richardson [mailto:bruce.richardson@intel.com]
> > > Sent: Wednesday, 20 December 2023 15.22
> > > 
> > > In order to remove more libraries from the mandatory list, we need to
> > > have support for optionally having a dependency from a driver or
> > > library
> > > to another driver or lib. This patch adds this support by adding a new
> > > optional_deps variable, the contents of which are added to the deps
> > > list
> > > if those optional dependencies are present in the build.
> > > 
> > > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > > ---
> > >  drivers/meson.build | 7 +++++++
> > >  lib/meson.build     | 7 +++++++
> > >  2 files changed, 14 insertions(+)
> > > 
> > > diff --git a/drivers/meson.build b/drivers/meson.build
> > > index 5ba534049a..af2d8da5a8 100644
> > > --- a/drivers/meson.build
> > > +++ b/drivers/meson.build
> > > @@ -127,6 +127,7 @@ foreach subpath:subdirs
> > >          includes = [include_directories(drv_path)]
> > >          # set up internal deps. Drivers can append/override as
> > > necessary
> > >          deps = std_deps
> > > +        optional_deps = []  
> > 
> > (I'm a meson noob, so please bear with my stupid questions.)
> > 
> > Is a separate "optional_deps" necessary? If a driver has any of these dependencies, why can it not just add them to the "deps" in the driver's meson.build file?
> > 
> > Ohhh... It's the other way around: The driver only depends on the other (optional) lib if that other lib is enabled! Correct?
> >   
> Yes, it's a shortcut to save an app having to manually check for its
> optional dependency itself in the meson.build file.
> 
> However, this is only really useful if it's an optional dependency where we
> just have #ifdefs in the C code for it. For the initial example I was
> thinking of to try and use it - meter lib in ethdev - something that simple
> is not enough. So long as there are extra C files or headers that need to
> be built when a dependency is present, we need to change the meson.build
> file to explicitly check anyway. Hence it's only an RFC until such time as
> we find a use-case or two that uses it.

Reasonable solution, but it hasn't been an issue for anything so far over a  year.
Marking it as "Awaiting upstream" since if a feature needs this, then use this.