kicad-developers team mailing list archive

[Dick Hollenbeck <dick@xxxxxxxxxxx>]

On 08/18/2014 06:47 PM, Wayne Stambaugh wrote:
> On 8/18/2014 6:45 PM, Brian Sidebotham wrote:
>> On 16 August 2014 17:44, Wayne Stambaugh <stambaughw@xxxxxxxxxxx> wrote:
>>> One of the tasks that I have committed to working on in the KiCad road
>>> map is to clean up the current mess we have created by allowing
>>> dependency libraries to be built as part of the KiCad source build.  The
>>> only exception I see for the time being is Boost.  Although I am have my
>>> reservations on that as well.  Why you ask?  I've spent several days
>>> trying to get KiCad to build on Windows using MSYS2 as my build
>>> environment and mingw64 as my target environment.  Every single library
>>> dependency with the exception of our custom Boost and avhttp (which
>>> could easily be build and installed using CMake) are already packaged
>>> for me.  However, the current KiCad build insists on downloading and
>>> building some libraries from source that are already installed.  This is
>>> silly.  I can resolve the issues by passing all of the
>>> PACKAGE_ROOT_PATHs when I run CMake but that is silly as well since my
>>> build environment already points the correct path.
>>>
>>> Originally I intended to create a separate project to build the KiCad
>>> library dependencies but I have since changed my mind.  I do *not* think
>>> it is asking too much of developers to learn how to build and/or install
>>> libraries on their preferred platform.  If as a developer you must have
>>> this done for you automatically, I am going to please ask that *you*
>>> create a separate platform specific build tool such as the excellent
>>> kicad-winbuilder that Brian has created.  This will significantly
>>> simplify the KiCad CMake files and eliminate the situation I described
>>> above.  My preference and goal is that the KiCad CMake files be used to
>>> build KiCad, not library dependencies.
>>>
>>> Initially, I plan to remove the dependencies that do not require any
>>> patching to build which currently are avhttp, swig, cairo, libpng,
>>> pixman, pcre, pkgconfig, and glew.  Then I will remove the dependencies
>>> with platform specific patches which are openssl, wxwidgets and
>>> wxpython.  Then I will try to figure out what to do with the problem
>>> child that is Boost.  I would also suggest that all platform specific
>>> library dependency build patches be remove as well leaving only the
>>> Boost patches that are required for all platforms (except the context
>>> switching patches).
>>>
>>> My goal here is not to step on anyone's toes it is to get our build
>>> system under control so that I can build *KiCad* rather than figure out
>>> how to get the dependencies to build or not as the platform dictates.  I
>>> expect our code to be well designed and I don't think expecting the same
>>> from our build system design is out of line.  If any one has major
>>> objections then we will have to figure something out because I am not
>>> going to continue to spend valuable time fighting our build tools to get
>>> them to properly use the dependencies already installed on my platform.
>>>
>>> Wayne
>>>
>>
>> Hi Wayne,
>>
>> I think that sounds like a sane decision, although of course KiCad
>> will be subject to the bugs in other libraries, but it's up to
>> distribution maintainers to sort those out in my opinion.
>>
>> I hope that we can use some sort of deprecation system, please mark a
>> lib as being deprecated so that I can sort out Winbuilder before the
>> CMake system is broke. It's much easier to work that way round as
>> opposed to a reactionary approach where we break everything and then
>> everyone has to fix their build before being able to do anything. I
>> will do the leg work to keep the Winbuilder people happy and do any
>> projects necessary to package dependencies required by it.
> 
> I will be making changes very slowly because I expect there to be some
> breakage with some of the FindFoo.cmake changes I'm going to make.  This
> will give everyone time to respond should I break anything.  I will be
> pushing hard against adding personal install paths to every
> FindFoo.cmake file.  CMake knows how to find the correct default paths
> on most platforms and I will always give the developer the chance to
> override those using either and environment variable or a definition on
> the CMake command line for custom library testing.  I will start out
> with the dependencies that don't require any patches to build and then
> address the more complex ones like wxWidgets and finish up with the
> nightmare that is Boost.  If I change more than one FindFoo.cmake every
> two weeks, that will be a blistering pace.  The goal is to get each
> dependency test working and properly designed before moving to the next one.
> 
> Cheers,
> 
> Wayne


The current thinking among those using CMake for several large projects is that it can
either find or it can build in the first pass.  It struggles with finding a prerequisite
and building the prerequisite as a fallback, then continuing to build that which does the
depending.  This is why when you dig into it, quite a number of larger projects have split
their CMake builders into two.  Sometimes this is done with or without a 3rd umbrella
builder on top of the two.

a) prerequisite builder
b) core project builder

c) optional umbrella builder to drive the above two.

If you can dovetail a) and b) into one builder, you very well may be the first person on
the planet to do it successfully, check with kitware.  For me, if I were going to invest
this kind of time, I would actually pay for a half hour of support from the authors of
CMake.  Just to get their view on this.  Then I would call them on the phone.


a) Is attractive, it is a mini form of open-embedded.  CMake could actually do a good job
of this.  And what is produced there is not specific to KiCad.  Therefore help may be
available within the open source community outside KiCad.

Open-embedded uses the notion of a staging area, and essentially there is no real
difference between building for the host machine vs. some other target with a different
instruction set.

Nobody maintaining large build systems is using "configure" directly, it is always wrapped
in something that records the exotic command line options.  Open-embedded calls these recipes.

As modular KiCad milestone C) rolls out, your biggest problem Wayne is going to be dealing
with python on Windows.  You will have to use python-a-mingw-us, and Brian's work there is
clearly leading edge.

Python-a-mingw-us is compiled using only a C compiler.  We've discovered that we can stray
from the same version of gcc as what python-a-mingw-us was build with, since it is a C
binary, within reason.

Once wxPython comes in, not so.  You should really use the same version of g++ to build
anything in the same process.  And the process will start with wxPython.

I am glad I am on linux.

You do not have to support the python project manager on Windows.  But I think that is
where much action will be since it will be easy to integrate 3rd party tools and you'll be
adding a whole lot of additional developers.


Dick