[OE-core] [PATCH] Yocto: Install full set of python modules in Qt SDK toolchain
Richard Purdie
richard.purdie at linuxfoundation.org
Fri Sep 19 07:25:12 UTC 2014
On Fri, 2014-09-19 at 09:14 +0200, Marek Vasut wrote:
> On Thursday, September 18, 2014 at 03:36:40 PM, Richard Purdie wrote:
> > On Thu, 2014-09-18 at 11:29 +0100, Laszlo Papp wrote:
> > > On Thu, Sep 18, 2014 at 10:30 AM, Marek Vasut <marex at denx.de> wrote:
> > > > On Thursday, September 18, 2014 at 11:23:13 AM, Laszlo Papp wrote:
> > > >> On Thu, Sep 18, 2014 at 10:16 AM, Marek Vasut <marex at denx.de> wrote:
> > > >> > On Thursday, September 18, 2014 at 11:11:45 AM, Laszlo Papp wrote:
> > > >> >> Come on, Yocto maintainers, please...
> > > >> >>
> > > >> >> Mark sent this change relatively long ago, and it is still in "new"
> > > >> >> state, sadly. The current SDK shipped _breaks_ any third-party
> > > >> >> software that uses standard python with regards to the libraries
> > > >> >> and all that.
> > > >> >>
> > > >> >> This is is slightly frustrating. We also face the same issue. :(
> > > >> >
> > > >> > Thanks for the reminder, new version which should fix the issues
> > > >> > with the previous one is on the ML now. You're all on CC.
> > > >>
> > > >> I do not think this is an explicit Qt issue, and hence fixing on that
> > > >> layer sounds like a weird approach. It seems to be a generic python
> > > >> issue and so, I think it should be addressed in its core. I opened a
> > > >> new report for this:
> > > >>
> > > >> https://bugzilla.yoctoproject.org/show_bug.cgi?id=6735
> > > >
> > > > Then you probably want to add nativesdk-packagegroup-python class into
> > > > the other toolchain recipes as well . Or even better, into the base
> > > > SDK toolchain class (is there one?)
> > >
> > > See, I do not understand why this "feature" was integrated the way it
> > > was. IMHO, it lacks any kind of reality. Has the SDK been ever tested
> > > against python based host development systems? My assumption is no. I
> > > do not think there is any sanity in using a _that_ stripped down
> > > version of python on desktop. It just really hurts the python users
> > > for the SDK, since there is no simple workaround and we cannot get the
> > > SDK to let the system python take precedence. This situation is awful
> > > in my opinion. Who is up for fixing this in the core? Let us have a
> > > _standard_ python shipped with the SDK by default or do not ship any
> > > at all.
> >
> > If you install nativesdk-python-modules into your SDK, you will get
> > python *and* all its modules installed. If you just install the python
> > core, you get a cut down python and need to install the modules you
> > need.
> >
> > The python SDK has been tested and used in a number of scenarios. Where
> > a full set of modules is needed, nativesdk-python-modules gets installed
> > and everyone is happy. If there is some problem with
> > nativesdk-python-modules, please let us know.
>
> git grep nativesdk-python-modules doesn't show any matches in
> git://git.yoctoproject.org/poky master . Do you mean the nativesdk-packagegroup-
> python I crafted or do you refer to something else please ?
>
> Also, shouldn't full python be installed into all the SDK toolchains ? I am for
> example unable to compile U-Boot 2014.10rc with the Yocto SDK toolchain anymore.
> The SDK is missing python modules and I cannot easily override the usage of
> python from the SDK . So I agree with Laszlo here, the SDK toolchain is somewhat
> unusable as it is.
I refer to the package I mentioned:
$ ls -la tmp/deploy/ipk/x86_64-nativesdk/nativesdk-python-modules*
-rw-r--r-- 2 richard richard 1512 Sep 18 17:09 tmp/deploy/ipk/x86_64-nativesdk/nativesdk-python-modules_2.7.3-r0.3.24_x86_64-nativesdk.ipk
See ${PN}-modules defined in python-2.7-manifest.inc.
I agreed there is a problem, I disagree somewhat about how it should be
fixed since you're just installing a set of modules which is defined as
those needed to run bitbake and I don't think this is what you actually
want. There is also the question of whether nativesdk-python should even
be in there...
Cheers,
Richard
More information about the Openembedded-core
mailing list