On 09/04/2014 07:05 PM, Mike Holmes wrote:



On 4 September 2014 11:54, Victor Kamensky <[email protected]
<mailto:[email protected]>> wrote:

    On 4 September 2014 08:11, Mike Holmes <[email protected]
    <mailto:[email protected]>> wrote:
     >
     >
     >
     > On 4 September 2014 11:01, Stuart Haslam <[email protected]
    <mailto:[email protected]>> wrote:
     >>
     >> On Thu, Sep 04, 2014 at 03:41:17PM +0100, Victor Kamensky wrote:
     >> > On 4 September 2014 03:10, Stuart Haslam
    <[email protected] <mailto:[email protected]>> wrote:
     >> > > On Wed, Sep 03, 2014 at 03:18:11AM +0100, Mike Holmes wrote:
     >> > >> Signed-off-by: Mike Holmes <[email protected]
    <mailto:[email protected]>>
     >> > >> ---
     >> > >>
     >> > >> This creates a new section in the documentaion to group
    everything
     >> > >> related to buffers. It appears to make things much more
    acessable
     >> > >> although it needs to have some real description added to
    the section.
     >> > >>
     >> > >
     >> > > Looks better to me. It also allows more freedom within the
     >> > > implementation
     >> > > to place things in different header files and still have the
     >> > > documentation
     >> > > look the same.
     >> >
     >> > If implementations will have freedom to place things in different
     >> > headers,
     >> > how one app source will work with all of them? Which header it
    would
     >> > include? I am not against fancy doxygen syntax, but mapping ODP
     >> > symbol to header file name should be normative part of ODP. Of
    course,
     >> > such mapping could be supported with set of implementation defined
     >> > headers that are internally included by one that constitute
    ODP API.
     >> >
     >> > Thanks,
     >> > Victor
     >> >
     >>
     >> This has come up before and I think the intention has always
    been that
     >> the application should just include odp.h and the implementation
    can do
     >> whatever it likes with header files beyond that (except perhaps that
     >> they need to be named odp_*.h).
     >>
     >> Personally though, I would rather the file names were fixed, and
    I did
     >> wonder when writing that comment whether this decision should be
     >> revisited.
     >>
     > I think in practice this will all work out nicely, but we need to
    work with
     > it a little, so we can work out the kinks.

    Not sure which way you argue :). Wrt of Stuart's only one odp.h few
    remarks:

    o in linux-generic odp_crypto.h and odp_rwlock.h are not included by
    odp.h. I guess it is not only me who did not know about one odp.h
    file rule.
    I think it should be fixed (i.e odp_crypto.h and odp_rwlock.h) must be
    added to odp.h.

Agree, I had to do this in my global_init fix, but it should just be
done. Personally I don't believe in odp.h, I think you should include
only what you need.

Mike, which disadvantages common odp.h has?
Was there some mail thread there pros and cons of common odp.h were
discussed.

_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to