Fwd: KF5 CMake usage question

Shaheed Haque srhaque at theiet.org
Mon May 22 18:54:39 BST 2017

On 21 May 2017 at 22:27, Aleix Pol <aleixpol at kde.org> wrote:
> On Sat, May 20, 2017 at 7:41 PM, Shaheed Haque <srhaque at theiet.org> wrote:
>> Actually, there is one thing about "target CMake"-based KF5 that I
>> don't quite understand: is there a way to get to the C++ compile flags
>> needed from CMake? That is, the modern equivalent of Foo_COMPILE_FLAGS
>> but for target Foo? Even if the general answer is "no", I'm interested
>> in at least the CMake variables/properties/commands needed to get to
>> "-fPIC" and "-std=gnu++14".
>> I'm aware of the target properties
>> POSITION_INDEPENDENT_CODE and CXX_STANDARD but none of these seem to
>> be set on targets I have tried.
>> Perhaps these are only set if somehow the compiler name etc. is specified?
>> Thanks, Shaheed
>> On 18 May 2017 at 18:04, Shaheed Haque <srhaque at theiet.org> wrote:
>>> Hi,
>>> On 18 May 2017 at 12:51, Andreas Hartmetz <ahartmetz at gmail.com> wrote:
>>>> On Samstag, 13. Mai 2017 23:48:33 CEST Shaheed Haque wrote:
>>>>> Hi,
>>>>> On 13 May 2017 at 22:04, Sven Brauch <mail at svenbrauch.de> wrote:
>>>>> > Hi,
>>>>> >
>>>>> > On 05/13/2017 06:06 PM, Shaheed Haque wrote:
>>>>> >> The printed output shows that the variable KF5KIO_INCLUDE_DIRS is
>>>>> >> not
>>>>> >> set. In poking around, I see references to a (new-to-me)
>>>>> >> target-based
>>>>> >
>>>>> >> system, and using that like this:
>>>>> > The question is, why do you need to do that?
>>>> The idea with the target based system aka "Modern CMake" is that you say
>>>> you want to compile against a library, and that is all you have to do. A
>>>> library requires you to add an include path for its own headers, include
>>>> paths for headers of one of its dependencies, and link against a bunch
>>>> of libraries? All covered by target properties.
>>>> If for some reason (e.g. handover to an external tool) you need those
>>>> properties, you can still query them. Under enforced names nonetheless,
>>>> unlike FOO_INCLUDE_DIR or was it FOO_INCLUDE_DIRS?
>>> Ack. The problem from the point of view of an automated tool which starts
>>> with a component called Foo arises ONLY because the target(s) of Foo are
>>> called FooFoo and FooBar. CMake does not - AFAICS - allow one to query Foo
>>> and somehow find FooFoo and FooBar inorder to look up FooFoo_INCLUDE_DIRS
>>> etc.
>>>>> I'm continuing to experiment with trying to build Python bindings for
>>>>> KF5. As part of that, the SIP tooling creates C++ wrapper code which
>>>>> must be compiled for each framework, and for that I need to know the
>>>>> header file directories. So far, I have simply been hardcoding the
>>>>> needed paths on my own system, but I now want to move to using
>>>>> standard CMake-based logic instead.
>>>>> (In some sense, this might be seen as similar to the stuff that was
>>>>> added to ECM, but I'm trying for a significantly more automated
>>>>> approach).
>>>>> Also, I am trying to feel my way towards a Pythonic build system for
>>>>> the KF5 bindings (as, roughly speaking, PyQt seems to be doing): in
>>>>> other words I'm interested in using CMake as a stepping stone, not the
>>>>> actual build system.
>>>> I would recommend against that unless you really need to have heavy
>>>> logic in the build system. A build system's main job is to "solve" a
>>>> dependency tree - that is the difference between a build system and a
>>>> script that runs the compiler. The dependency tree enables cheap
>>>> incremental builds and correct parallel builds. Maybe not that important
>>>> for bindings, admittedly.
>>>> Your advantage from using Python must be larger than the overhead from
>>>> doing your own dependency resolution plus the overhead from the CMake-
>>>> Python interfacing plus the build-related facilities that CMake has and
>>>> Python doesn't. Or were you considering scons?
>>>> PyQt may have chosen Python because qmake sucks, and it needs Python
>>>> anyway, so it avoids any extra dependencies. I know from experience that
>>>> you really want to avoid extra dependencies in commercial products.
>>> /me nods vigourosly.
>>> I'm not (yet) familair with all the intricacies of the Python build system
>>> (or CMake for that matter!), but I do see that PyQt has to work quite hard
>>> to keep its build system working as a Python user might expect. Further, the
>>> system I am seeking to build has to support more than KF5 (or even KDE). So,
>>> roughly speaking, the split I am going for is:
>>> - Keep all platform and system independent code in Python
>>> - Isolate all platform and system independent logic in CMake
>>> As I say, I am feeling my way a bit here, but this seems like a
>>> philosophically justifiable separation. Oh, and to solve the problem of
>>> finding the targets, I resorted to parsing the CMake files (!!). I can live
>>> with that hack precisely because by having the split, users of this code who
>>> are not using it against KF5 will need to replace this CMake part with their
>>> own anyway.
>>> (At this point, abstracting CMake away entirely is a minor detail).
>>> Thanks for the helpful remarks.
>>> Shaheed
>>>>> Thus, I'm after the moral equivalents of:
>>>>> Thanks, Shaheed
>>>>> > The usual way is to simply call
>>>>> >
>>>>> > target_link_libraries(mybinary KF5::KIOCore)
>>>>> >
>>>>> > and include paths etc. will be set up for your target automatically.
>>>>> >
>>>>> > Best,
>>>>> > Sven
>>>> Cheers,
>>>> Andreas M9
> You can easily see how it works in extra-cmake-modules code.

I don't think it is quite so simple. First, for "-std=gnu++14",the ECM
code just hardcodes it (with a comment that hints at difficulties
extracting this value). Second, it grabs INTERFACE_COMPILE_DEFINITIONS
and INTERFACE_COMPILE_OPTIONS of which the latter seems to be needed
in order to get stuff from come KF5 cmake file which set -fexception
(in some case, for example).

But what sets "-fPIC"? It seems to be inherited in some way from
Qt5::Core INTERFACE_COMPILE_OPTIONS, but how can I get to that

Thanks, Shaheed

> Aleix

More information about the kde-core-devel mailing list