conf/machine/intel-corei7-64.conf includes ${XSERVER_X86_MATROX_MGA} in ${XSERVER}, and conf/machine/include/meta-intel.inc sets {XSERVER_X86_MATROX_MGA} to "xf86-video-mga". xf86-video-mga used to be in OE-core, but now it is in meta-oe. The result is that building packagegroup-core-x11[-xserver].bb fails when using meta-intel without meta-oe, which probably isn't intended? The fix is trivial (remove all references to XSERVER_X86_MATROX_MGA and xf86-video-mga); the bigger question is why that hasn't been caught by tests for meta-intel. Does testing have to be enhanced? We ran into this in refkit, which has meta-oe, but carefully controls which recipes from it get used - see https://github.com/intel/intel-iot-refkit/pull/258
(In reply to comment #0) > xf86-video-mga used to be in OE-core, but now it is in meta-oe. Close, but not quite correct. It's still in meta-intel, but when meta-oe is active, the copy from meta-oe is used. Is that how it should be? > The fix is trivial (remove all references to XSERVER_X86_MATROX_MGA and > xf86-video-mga); the bigger question is why that hasn't been caught by tests > for meta-intel. Does testing have to be enhanced? Nope. meta-intel without meta-oe works fine.
Could this have to do with Layer Priority? meta-intel includes a full mga recipe, for 1.6,4, which is older than the 1.6.5 version in meta-oe. I was not aware that it moved to meta-oe. Does it even make sense to have the extra XSERVER video drivers ?
(In reply to comment #2) > Could this have to do with Layer Priority? meta-intel includes a full mga > recipe, for 1.6,4, which is older than the 1.6.5 version in meta-oe. Yes, it is because meta-intel has a lower priority (5) than meta-oe (6). > I was not aware that it moved to meta-oe. commit 29e111093bc97806ea5e96d1526fa93b055fac18 Author: Jackie Huang <jackie.huang@windriver.com> Date: Tue May 23 15:01:41 2017 +0800 xf86-video-mga: add new recipe mga is an Xorg driver for Matrox video cards Signed-off-by: Jackie Huang <jackie.huang@windriver.com> Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com> > Does it even make sense to have the extra XSERVER video drivers ? I think it makes sense to make it configurable which drivers get installed. That's not a choice that should be hard-coded in OE-core. I'm less sure about meta-intel carrying its own copy of xf86-video-mga, and how that is meant to play together with other layers also having a copy. If the goal is that for Intel hardware the driver gets used that ships with meta-intel and that presumably was tested as part of meta-intel releases, then the recipe might have to be renamed to ensure that the right driver gets picked depending on the MACHINE and thus XSERVER setting. Alternatively one could raise the priority of meta-intel, but then one gets into situations where one cannot use the other driver even if one wants to. If the driver does not get tested, then perhaps it should get removed entirely from meta-intel?
Upgraded the meta-intel recipe to use 1.6.5 xserver video mga. http://git.yoctoproject.org/cgit/cgit.cgi/meta-intel/commit/?id=92ce2459bc82b9ecdf9623c3008b7a8f427506dd
(In reply to comment #4) > Upgraded the meta-intel recipe to use 1.6.5 xserver video mga. > http://git.yoctoproject.org/cgit/cgit.cgi/meta-intel/commit/ > ?id=92ce2459bc82b9ecdf9623c3008b7a8f427506dd How does that fix the problem? Because of layer priorities, meta-oe's copy of xf86-video-mga is still going to get used instead of the recipe in meta-intel.
So the only way to fix this would be to either lower meta-oe's priority or raise meta-intel's, if we move meta-intel to 7 (above meta-oe) what other fallout/issues will be caused by other layers. Currently the 2 recipes are identical, I know that could change in the future. MGA is legacy graphics hardware (circa 2006), so does it continue to make sense to carry it and maybe the AST driver in meta-intel any longer? Both MGA and AST are used with non-Intel hardware, so having them in meta-intel may not makes sense longer term.
Long term I think meta-oe should have the default/normal/average priority, as it's meant to be a layer of extra recipes. I suspect it's too late in the cycle to do this in meta-oe though.
(In reply to comment #6) > MGA is legacy graphics hardware (circa 2006), so does it continue to make > sense to carry it and maybe the AST driver in meta-intel any longer? Was that a question for me? I don't have an opinion about this.
Given the information I have currently, I will be removing MGA and AST from meta-intel completely and from the XSERVER config early in 2.5
Recipe removed here (thanks Anuj): https://git.yoctoproject.org/cgit/cgit.cgi/meta-intel/commit/?id=7bf03f7612834142355189c46ccdb4b309acc58a