10/01/2019 11:02, Bruce Richardson:
> On Thu, Jan 10, 2019 at 09:52:43AM +0100, Thomas Monjalon wrote:
> > 10/01/2019 09:45, Burakov, Anatoly:
> > > On 09-Jan-19 9:36 PM, Thomas Monjalon wrote:
> > > > 09/01/2019 21:51, Bruce Richardson:
> > > >> One other point I'd make is that having compat as a separate library
> > > >> seems
> > > >> a little like overkill to me - it's just one header file! Is there a
> > > >> reason
> > > >> why we don't just move it into the EAL where everyone can use it
> > > >> without
> > > >> having to specify another dependency?
> > > >
> > > > +1 for moving compat into EAL.
> > >
> > > Presumably the reason to keep it separate would be to enable depending
> > > on it for libraries that EAL itself depends on (i.e. kvargs).
> >
> > Yes, forgot this one.
>
> I also forgot that. However, that does bring up the thought that for some
> of our headers we may want them available for all libraries even before EAL
> is compiled up. [I think this hits on the idea Thomas was discussing of
> splitting EAL up into two, where we have a core part that is always
> compiled first and has no dependencies followed by another which is built
> later and can have dependencies.]
>
> For this specific issue, my thought is that we should indeed move compat.h
> into EAL, but also add the EAL common/include/ path to the global include
> path for the project, so all EAL headers (e.g. including those with defines
> for cacheline size, and other common macros) are just always available.
> Libs like kvargs and others should be able to use RTE_DIM etc. for
> consistency with the rest of DPDK.
You are describing a cheap split :)
It would bring some definitions but not some non-inline functions.
I still think we should split in 2 libraries:
- EAL as "Environment Abstraction Layer", i.e. basic stuffs
- Initialization + core features (bus/dev and mem)
We could even move bus/dev and mem in separate libraries.
