Skip to contentSkip to conversation

Add nativesdk tools for Qt5 SDK

PATCH In review Dominik Holland 2 participants 3 messages Dec 2, 2013

lore
Share & tools
lore
#0 lore raw

cover Add nativesdk tools for Qt5 SDK

Hi all,

i'm the author of the merge-request on github.

The patchset from Denys looks nice. Especially because it cross compiles the host tools for the SDK_MACHINE which i forgot todo in my patchset.

I think the best way would be a combined version of both patchsets.

Because some of you think my patchset is too complicated i want to highlight why i added the complicated creation of the qt.conf and the additional patching of the mkspec files etc.

I didn't tested Denys version but i think it only works if you source the environment scripts of the SDK before.

As a Qt developer it's the defacto standard to get a SDK which works out of the box for your projects by just invoking qmake and no need to source a script before which changes your environment.

To get this done you need to patch the mkspec folder and the qt.conf to use relative paths instead the hardcoded paths which where created during the yocto build.

In addition i created a linux-oe-sdk-g++ which contains the real values of the environment variables but without --sysroot. This will be set by qt itself by using the CROSS_COMPILE prefix.

I hope you understand that this is really needed if you want to get a good qt5 integration into a OE SDK.

Best Reagrds

Dominik

P.S. If you want to use the qmake inside qtcreator you would need to set all the env variable of the environment script inside qtcreator which is really inconvenient

#1 lore raw

Sorry for top-posting, but I wanted to reply in general and couldn't find the right place for inlining...

Anyway, the below assumptions aren't exactly correct. And the main one being - you don't really need to source the environment-setup script. But you'd need a different and much simpler qt.conf, which I populate as part of the meta-toolchain recipe (still pending submission though). And you can use relative paths in that generated qt.conf so qmake can be used w/o sourcing the environment-setup script. But even with absolute paths those will be fixed automatically by the SDK setup script at the install time...

The thing is, you cannot re-use the same qt.conf that you generate for the nativesdk tools, because that one is for cross-compiling for SDK. Like you mentioned, it cross-compiles the tools for SDK_MACHINE, which is different from target MACHINE as well as the build host. When the tools are used, you need a different qt.conf with a makespec to run on SDK_MACHINE and build for target MACHINE.

Also, I wanted the things to be as simple as possible, so I decided to re-use linux-oe-g++ makespec, as it is already there and we've been using it successfully for a long time with Qt4. The only thing is, existing linux-oe-g++ relies on the OE-specific variables, because that was the way OE SDK always worked.

Signature · 1 line
Denys
Quoted 44 lines · Dominik Holland

Hi all,

i'm the author of the merge-request on github.

The patchset from Denys looks nice. Especially because it cross compiles the host tools for the SDK_MACHINE which i forgot todo in my patchset.

I think the best way would be a combined version of both patchsets.

Because some of you think my patchset is too complicated i want to highlight why i added the complicated creation of the qt.conf and the additional patching of the mkspec files etc.

I didn't tested Denys version but i think it only works if you source the environment scripts of the SDK before.

As a Qt developer it's the defacto standard to get a SDK which works out of the box for your projects by just invoking qmake and no need to source a script before which changes your environment.

To get this done you need to patch the mkspec folder and the qt.conf to use relative paths instead the hardcoded paths which where created during the yocto build.

In addition i created a linux-oe-sdk-g++ which contains the real values of the environment variables but without --sysroot. This will be set by qt itself by using the CROSS_COMPILE prefix.

I hope you understand that this is really needed if you want to get a good qt5 integration into a OE SDK.

Best Reagrds

Dominik

P.S. If you want to use the qmake inside qtcreator you would need to set all the env variable of the environment script inside qtcreator which is really inconvenient
_______________________________________________
Openembedded-devel mailing list [email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-devel

#2 lore raw

Hi,

Quoted 10 lines · Denys Dmytriyenko

Sorry for top-posting, but I wanted to reply in general and couldn't find the right place for inlining...

Anyway, the below assumptions aren't exactly correct. And the main one being - you don't really need to source the environment-setup script. But you'd need a different and much simpler qt.conf, which I populate as part of the meta-toolchain recipe (still pending submission though). And you can use relative paths in that generated qt.conf so qmake can be used w/o sourcing the environment-setup script. But even with absolute paths those will be fixed automatically by the SDK setup script at the install time…

ok great, if you have a much simpler qt.conf which is populated and works. I'm find with that ;-)

Quoted 7 lines

The thing is, you cannot re-use the same qt.conf that you generate for the nativesdk tools, because that one is for cross-compiling for SDK. Like you mentioned, it cross-compiles the tools for SDK_MACHINE, which is different from target MACHINE as well as the build host. When the tools are used, you need a different qt.conf with a makespec to run on SDK_MACHINE and build for target MACHINE.

I know, and i didn't said anything that i wanted to do that.

Quoted 6 lines

Also, I wanted the things to be as simple as possible, so I decided to re-use linux-oe-g++ makespec, as it is already there and we've been using it successfully for a long time with Qt4. The only thing is, existing linux-oe-g++ relies on the OE-specific variables, because that was the way OE SDK always worked.

I'm also fine with that.

What i'm really up to is to fix the hardcoded paths in mkspecs/qdevice.pri and mkspecs/qconfig.pri

Don't get me wrong i really appreciate your work, i only want to make sure that the result works in every way and a qt-developer doesn't have to do any additional setup before he can use the SDK.

Have fun!
Dominik

Quoted 54 lines
Signature · 2 lines
-- 
Denys
Quoted 37 lines · Dominik Holland

Hi all,

i'm the author of the merge-request on github.

The patchset from Denys looks nice. Especially because it cross compiles the host tools for the SDK_MACHINE which i forgot todo in my patchset.

I think the best way would be a combined version of both patchsets.

Because some of you think my patchset is too complicated i want to highlight why i added the complicated creation of the qt.conf and the additional patching of the mkspec files etc.

I didn't tested Denys version but i think it only works if you source the environment scripts of the SDK before.

As a Qt developer it's the defacto standard to get a SDK which works out of the box for your projects by just invoking qmake and no need to source a script before which changes your environment.

To get this done you need to patch the mkspec folder and the qt.conf to use relative paths instead the hardcoded paths which where created during the yocto build.

In addition i created a linux-oe-sdk-g++ which contains the real values of the environment variables but without --sysroot. This will be set by qt itself by using the CROSS_COMPILE prefix.

I hope you understand that this is really needed if you want to get a good qt5 integration into a OE SDK.

Best Reagrds

Dominik

Quoted 7 lines

P.S. If you want to use the qmake inside qtcreator you would need to set all the env variable of the environment script inside qtcreator which is really inconvenient
_______________________________________________
Openembedded-devel mailing list [email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-devel

_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-devel