[dpdk-dev] [PATCH v12 01/14] build: alias default build as generic

Bruce Richardson bruce.richardson at intel.com
Tue Nov 17 10:15:18 CET 2020


On Tue, Nov 17, 2020 at 08:49:45AM +0100, Thomas Monjalon wrote:
> 17/11/2020 03:46, Honnappa Nagarahalli:
> > <snip>
> > 
> > > 
> > > 16/11/2020 17:16, Bruce Richardson:
> > > > On Mon, Nov 16, 2020 at 03:50:31PM +0000, Juraj Linkeš wrote:
> > > > > From: Thomas Monjalon <thomas at monjalon.net>
> > > > > > 13/11/2020 15:31, Juraj Linkeš:
> > > > > > > The current machine='default' build name is not descriptive. The
> > > > > > > actual default build is machine='native'. Add an alternative
> > > > > > > string which does the same build and better describes what we're
> > > building:
> > > > > > > machine='generic'. Leave machine='default' for backwards
> > > compatibility.
> > > > > >
> > > > > > What?
> > > > > >
> > > > > > "generic" means... nothing.
> > > > > >
> > > > >
> > > > > An absence of anything means nothing. Generic means "characteristic of
> > > or relating to a class or group of things; not specific", which is pretty much
> > > what we're looking for.
> > > > >
> > > > > > "default" should be the most common set of options to make a build
> > > > > > work everywhere.
> > > > >
> > > > > What we want is a value of machine that would "be the most common
> > > set of options to make a build work everywhere" and using the above
> > > definition of generic, it fits very well.
> > > > > The reason I said the actual default build is machine='native' is because
> > > that's how the machine option is defined in meson_options.txt. It follows
> > > from what default actually means - "a preselected option adopted by a
> > > computer program or other mechanism when no alternative is specified by
> > > the user or programmer". Default then means no user input, which means
> > > machine='native', which means the default build is the default build.
> > > > >
> > > > > What ""default" should mean" looks like an attempt at redefining what
> > > the word actually means and leads to confusion, in my experience. Hence an
> > > attempt to remove the potential ambiguity.
> > > > >
> > > >
> > > > I would tend to agree that "generic" is probably a better term than
> > > > "default" for what we use it for here in the config.
> > > 
> > > In the past, we had a different definition with make config.
> > > I am just trying to be consistent.
> > > Even with meson, default means "minimal CPU instructions".
> > > 
> > > Example in devtools/test-meson-builds.sh:
> > > "test compilation with minimal x86 instruction set"
> > > is called build-x86-default.
> > > 
> > > In config/meson.build:
> > > "
> > > machine type 'default' is special, it defaults to the per arch agreed common
> > > minimal baseline needed for DPDK.
> > > That might not be the most optimized, but the most portable version while
> > > still being able to support the CPU features required for DPDK.
> > > This can be bumped up by the DPDK project, but it can never be an invariant
> > > like 'native'
> > > "
> > > 
> > > So, why this definition is called "generic" in meson Arm config?
> > The explanation above is for a build type 'default'. Whereas meson by default builds for build type 'native'. Also when you look at the config/arm/meson.build the word 'default' was used where it was not related to the build type default. It created lot of confusion.
> > 
> > From the dictionary 'default' - "a preselected option adopted by a computer program or other mechanism when no alternative is specified by the user or programmer." But, if one had to do build of type default, they have to mention -Dmachine=default. If nothing is mentioned, it is a build type 'native', which does not go along with the definition of 'default'.
> > 
> > But for 'generic' - "characteristic of or relating to a class or group of things; not specific". IMO, it better suits the explanation you have provided above. So, separating this machine type to 'generic' to cover the same definition makes more sense.
> > 
> > However, 'default' is still supported for backward compatibility.
> 
> So? Are you going to change the DPDK definitions we had for years?
> 

I think we should, or at least support "generic" alongside it. Using the term
"default" for something that is not the default is confusing. It's also not
a good description of what the result is, since it's a generic binary for
the architecture, rather than a "default" one for the architecture.

/Bruce


More information about the dev mailing list