[Mesa-dev] Meson mesademos (Was: [RFC libdrm 0/2] Replace the build system with meson)
Jose Fonseca
jfonseca at vmware.com
Thu Mar 30 22:19:31 UTC 2017
On 30/03/17 19:52, Dylan Baker wrote:
> Quoting Jose Fonseca (2017-03-29 15:27:58)
>> On 28/03/17 22:37, Dylan Baker wrote:
>>> Quoting Jose Fonseca (2017-03-28 13:45:57)
>>>> On 28/03/17 21:32, Dylan Baker wrote:
>>>>> Quoting Jose Fonseca (2017-03-28 09:19:48)
>>>>>> On 28/03/17 00:12, Dylan Baker wrote:
>>>>>>> Quoting Jose Fonseca (2017-03-27 09:58:59)
>>>>>>>> On 27/03/17 17:42, Dylan Baker wrote:
>>>>>>>>> Quoting Jose Fonseca (2017-03-27 09:31:04)
>>>>>>>>>> On 27/03/17 17:24, Dylan Baker wrote:
>>>>>>>>>>> Quoting Jose Fonseca (2017-03-26 14:53:50)
>>>>>>>>>>>> I've pushed the branch to mesa/demos, so we can all collaborate without
>>>>>>>>>>>> wasting time crossporting patches between private branches.
>>>>>>>>>>>>
>>>>>>>>>>>> https://cgit.freedesktop.org/mesa/demos/commit/?h=meson
>>>>>>>>>>>>
>>>>>>>>>>>> Unfortunately, I couldn't actually go very far until I hit a wall, as
>>>>>>>>>>>> you can see in the last commit message.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> The issue is that Windows has no standard paths for dependencies
>>>>>>>>>>>> includes/libraries (like /usr/include or /usr/lib), nor standard tool
>>>>>>>>>>>> for dependencies (no pkgconfig). But it seems that Meson presumes any
>>>>>>>>>>>> unknown dependency can be resolved with pkgconfig.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> The question is: how do I tell Meson where the GLEW headers/library for
>>>>>>>>>>>> MinGW are supposed to be found?
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> I know one solution might be Meson Wraps. Is that the only way?
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> CMake makes it very easy to do it (via Cache files as explained in my
>>>>>>>>>>>> commit message.) Is there a way to achieve the same, perhaps via
>>>>>>>>>>>> cross_file properties or something like that?
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Jose
>>>>>>>>>>>
>>>>>>>>>>> I think there are two ways you could solve this:
>>>>>>>>>>>
>>>>>>>>>>> Wraps are probably the most generically correct method; what I mean by that is
>>>>>>>>>>> that a proper wrap would solve the problem for everyone, on every operating
>>>>>>>>>>> system, forever.
>>>>>>>>>>
>>>>>>>>>> Yeah, that sounded a good solution, particularly for windows where's so
>>>>>>>>>> much easier to just build the dependencies as a subproject rather than
>>>>>>>>>> fetch dependencies from somewhere, since MSVC RT versions have to match
>>>>>>>>>> and so.
>>>>>>>>>>
>>>>>>>>>> > That said, I took a look at GLEW and it doesn't look like a
>>>>>>>>>>> straightforward project to port to meson, since it uses a huge pile of gnu
>>>>>>>>>>> makefiles for compilation, without any autoconf/cmake/etc. I still might take a
>>>>>>>>>>> swing at it since I want to know how hard it would be to write a wrap file for
>>>>>>>>>>> something like GLEW (and it would probably be a pretty useful project to wrap)
>>>>>>>>>>> where a meson build system is likely never going to go upstream.
>>>>>>>>>>
>>>>>>>>>> BTW, regarding GLEW, some time ago I actually prototyped using GLAD
>>>>>>>>>> instead of GLEW for mesademos:
>>>>>>>>>>
>>>>>>>>>> https://cgit.freedesktop.org/~jrfonseca/mesademos/log/?h=glad
>>>>>>>>>>
>>>>>>>>>> I find GLAD much nicer that GLEW: it's easier to build, it uses upstream
>>>>>>>>>> XML files, it supports EGL, and it's easy to bundle.
>>>>>>>>>>
>>>>>>>>>> Maybe we could migrate mesademos to GLAD as part of this work instead of
>>>>>>>>>> trying to get glew "mesonfied".
>>>>>>>>>>
>>>>>>>>>>> The other option I think you can use use is cross properties[1], which I believe
>>>>>>>>>>> is the closest thing meson has to cmake's cache files.
>>>>>>>>>>>
>>>>>>>>>>> I've pushed a couple of commits, the last one implements the cross properties
>>>>>>>>>>> idea, which gets the build farther, but then it can't find the glut headers,
>>>>>>>>>>> and I don't understand why, since "cc.has_header('GL/glut')" returns true. I
>>>>>>>>>>> still think that wraps are a better plan, but I'll have to spend some time today
>>>>>>>>>>> working on a glew wrap.
>>>>>>>>>>>
>>>>>>>>>>> [1] https://github.com/mesonbuild/meson/wiki/Cross-compilation (at the bottom
>>>>>>>>>>> under the heading "Custom Data")
>>>>>>>>>>
>>>>>>>>>> I'm running out of time today, but I'll try to take a look tomorrow.
>>>>>>>>>>
>>>>>>>>>> Jose
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I'd had a similar thought, but thought of libpeoxy? It supports the platforms we
>>>>>>>>> want, and already has a meson build system that works for windows.
>>>>>>>>
>>>>>>>> I have no experience with libepoxy. I know GLAD is really easy to
>>>>>>>> understand, use and integrate. It's completly agnostic to toolkits like
>>>>>>>> GLUT/GLFW/etc doesn't try to alias equivalent entrypoints, or anything
>>>>>>>> smart like libepoxy.
>>>>>>>>
>>>>>>>> In particular I don't fully understand libepoxy behavior regarding
>>>>>>>> wglMakeCurrent is, and whether that will create problems with GLUT,
>>>>>>>> since GLUT will call wglMakeCurrent..
>>>>>>>>
>>>>>>>>
>>>>>>>> Jose
>>>>>>>
>>>>>>> Okay, I have libepoxy working for windows. I also got libepoxy working as a
>>>>>>> subproject, but it took a bit of hacking on their build system (there's
>>>>>>> some things they're doing that make them non-subproject safe, I'll send patches
>>>>>>> and work that out with them.
>>>>>>>
>>>>>>> https://github.com/dcbaker/libepoxy.git fix-suproject
>>>>>>
>>>>>> Thanks.
>>>>>>
>>>>>> GLEW is not the only one case though. There's also FREEGLUT. So we
>>>>>> can't really avoid the problem of external windows binaries/subprojects.
>>>>>>
>>>>>> So I've been thinking, and I suspect is better if first get things
>>>>>> working with binary GLEW / FREGLUT projects, then try the glew ->
>>>>>> libepoxy in a 2nd step, so there's less to take in to merge meson into
>>>>>> master.
>>>>>>
>>>>>>> Clone that repo into $mesa-demos-root/subprojects and things should just work,
>>>>>>> or mostly work. I got epoxy compiling, but ran into some issues in the mingw glu
>>>>>>> header.
>>>>>>>
>>>>>>> Dylan
>>>>>>
>>>>>> I'm pretty sure the problem with MinGW glu is the lack of windows.h. We
>>>>>> need to do the same as CMakeLists.txt snippet quoted below.
>>>>>>
>>>>>> I'm running out of time today, but I'll look into porting this over to
>>>>>> meson tomorrow if you don't beat me to it.
>>>>>>
>>>>>> Jose
>>>>>>
>>>>>>
>>>>>>
>>>>>> if (WIN32)
>>>>>> # Nobody likes to include windows.h:
>>>>>> # - Microsoft's GL/gl.h header depends on windows.h but doesn't
>>>>>> include it;
>>>>>> # - GLEW temporarily defines the necessary defines but
>>>>>> undefines them later
>>>>>> # - certain GLUT distributions don't include it;
>>>>>> # - most of our programs are meant to be portable so don't
>>>>>> include it.
>>>>>> #
>>>>>> # We could try to replicate the windows.h definitions required by
>>>>>> # GL/gl.h, but the build time savings don't compensate the constant
>>>>>> # headaches that brings, so instead we force windows.h to be
>>>>>> included
>>>>>> # on every file.
>>>>>> if (MSVC)
>>>>>> add_definitions (-FIwindows.h)
>>>>>> else (MSVC)
>>>>>> add_definitions (--include windows.h)
>>>>>> endif (MSVC)
>>>>>>
>>>>>> # Don't define min/max macros
>>>>>> add_definitions (-DNOMINMAX)
>>>>>>
>>>>>> # MSVC & MinGW only define & use APIENTRY
>>>>>> add_definitions (-DGLAPIENTRY=__stdcall)
>>>>>>
>>>>>> link_libraries (winmm)
>>>>>> endif (WIN32)
>>>>>>
>>>>>
>>>>> Okay, I got things compiling and sorta working with mingw (tested with wine,
>>>>> so...) and libepoxy. You might be right that the epoxy work is something that we
>>>>> should do separately, I may try to write a freeglut wrap, it looks a bit more
>>>>> straight forward than glew.
>>>>
>>>> I've been pondering on this further, and I wonder if a good compromise a
>>>> subproject for glew/freeglut, that instead of building those said
>>>> dependencies, would simply fetch and unpack the files in a build subdir,
>>>> and would setup the meson dependency object (ie, path to includes, path
>>>> to libs.)
>>>>
>>>> For example, the freeglut project would download the windows binaries,
>>>> 32 or 64 bits based on the target build, perhaps checksum them, unpack,
>>>> etc.
>>>>
>>>> This would have the feel of a wrap, without the headaches of actually
>>>> porting the dependency to meson. (Nothing prevent us of doing that in
>>>> the long term thought. But it would simplify things in the short term.)
>>>>
>>>> We could host the binaries somewhere in freedesktop.org too for
>>>> protecting against broken urls etc.
>>>>
>>>> Jose
>>
>> I found
>> https://github.com/mesonbuild/meson/wiki/Shipping-prebuilt-binaries-as-wraps
>> which describes how to this precisely this.
>>
>>>
>>> That might be a viable option too. I'm still going to have a go at freeglut, it
>>> should be pretty easy to do, and getting it in the wrapdb would probably be
>>> useful.
>>>
>>> Dylan
>>>
>>
>> Sure.
>>
>> Jose
>
> That does look incredibly useful, especially for mesa. There are some
> dependencies that I have no interest in porting to meson (LLVM), I did get
> freeglut ported to meson (It was pretty easy), and tested it for mingw cross
> compiling. It's pushed upstream.
>
> Libepoxy also fixed their build upstream to work as a dependency, so I've added
> a wrap that just clones upstream using git, and we can migrate that to not git
> when they make another release, or we can package glew as a binary wrap, and
> deal with using a epoxy or glad or something later.
>
> Upstream is also following this thread and has opened this bug:
> https://github.com/mesonbuild/meson/issues/1524
>
> So that may simplify some things even further.
>
> Dylan
>
Cool. BTW, another alternative (for things like LLVM) would be to
chain build systems (ie, have a wrap that builds LLVM invoking CMake)
Jose
More information about the mesa-dev
mailing list