Re: [meta-oe][PATCH] qt-creator: avoid conflicts with meta-qt5's qt5-creator

[Martin Jansa <martin.jansa@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3142 bytes --]

On Thu, Mar 05, 2015 at 09:57:37PM +0100, Andreas Müller wrote:
> On Thu, Mar 5, 2015 at 7:01 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
> > On Sun, Mar 01, 2015 at 11:23:44PM +0100, Andreas Müller wrote:
> >> We avoid conflicts by installing no files to sysroot. This causes no fallout
> >> because nothing depends on qt5-creator (if something included later depending
> >> on qt-creator it should depend on qt5-creator).
> >> Note that
> >>
> >> | WARNING: QA Issue: qt5-creator rdepends on qt-creator, but it isn't a build dependency? [build-deps]
> >>
> >> is a false positve because the names of the libraries are same as for
> >> qt5-creator (see test below).
> >
> > This isn't enough to resolve the warning (but it resolves the sysroot
> > conflict which is good).
> >
> > qt-creator is still recorded as runtime provider for couple of
> > libraries, we can list them all in PRIVATE_LIBS (if we can assume that
> > nothing will link against them - which is already assumed by not staging
> > them).
> >
> > Here is the list from qt5-creator log.do_package:
> >
> > $ grep "requires package qt-creator" log.do_package
> > DEBUG: qt5-creator: Dependency libCore.so requires package qt-creator (used by files: /home/jenkins/oe/world/shr-core/tmp-glibc/work/armv5e-oe-linux-gnueabi/qt5-creator/3.3.1-r0/packages-split/qt5-creator/usr/lib/qt5/qtcreator/plugins/libQmlJSTools.so)
> Very interesting:
> 
> * qt5-creator complains for libs it has installed itself (see qt5 in path)
> * how should someone link against libraries that are not in sysroot?

See the code for shlibs providers in package.bbclass, this warning and
this whole issue originates there.

shlibs providers record all installed libraries (not staged in sysroot),
so that it can automatically add RDEPENDS in case some other package
will require already found library.

This happens with qt*-creators, there are plans to improve it by
including the whole path, but that doesn't work yet with current
package.bbclass.

> * during my tests for sysroot patch I have build qt-creator from
> scratch and then an image with qt5-creator included. If this warning
> would be true the image creation would have failed because qt5-creator
> RCONFLICTS qt-creator (error occured when starting the last patch)
> 
> Problem seems that that the instance creating (have not looked
> further) these warnings just checks the name of the library without
> path. Library names are same for both versions of qt-creator.
> 
> We could try it with PRIVATE_LIBS but I'd suggest to simply ignore
> these false warnings.

I was suggesting PRIVATE_LIBS because it sort of matches with skipping
the staging of the libraries.

The warning is still valid in cases when someone builds qt-creator
first, then qt5-creator which will automatically get RDEPENDS on
qt-creator, so installing qt5-creator from package feed will bring both
of them :(.

If we set PRIVATE_LIBS in qt-creator, it won't ever be registered as
shlib provider, so qt5-creator won't get the RDEPENDS.

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 188 bytes --]