[cmake-developers] -GNinja on Windows

Peter Kümmel syntheticpp at gmx.net
Wed Jun 6 11:50:41 EDT 2012


On 06.06.2012 16:27, Bill Hoffman wrote:
>>
>> Maybe there is a better solution, which would automatically fix a lot of
>> unit tests.
> Can you bring that up on the ninja mailing list?

I tried it with not using slashes:

     https://github.com/syntheticpp/ninja/commit/262b7b951135eac51023512d2f788727828bac43
     https://github.com/syntheticpp/CMake/commit/c95404d65229d1dd37f0579afb29e27f95501e06

but it doesn't help much. I have the impression it's a multi threading problem.
Linking is only possible with -j1.

> Is that a patch that could go upstream ninja?

Yes, the link points to a pull request.


>
> Sorry, no can do.  If it is there, people will expect it to work for

Yes, you are right. More testing with mingw shows that it is completely unusable.

> their project.  Right now it is not working, so I don't want to enable
> it.  The only feedback we will get, is these things are broken.  We know
> what is broken, what more feedback do we need?
>
> To move forward on windows, I will change the dashboard to use ninja
> master.  Right now it is still using your branch on git and is failing a
> bunch of tests. Also, it sounds like we took a step backwards with MinGW
> and the response files.  Maybe until we figure that out we should
> disable response files for MinGW?

ninja/master is not usable with mingw regardless what we do. And it has
not much to do with the rsp files, most unit tests don't use a response file.

Maybe I should switch back to my branch, master is a big mess for mingw.

Did you ever run the unit tests before with mingw? Arn't they all broken
because of the missing CMAKE_RC_COMPILER?

>
> Perhaps we can try one of the forks that has depend information for

Seems we are forced to, nothing happen there.

> MSVC?  If one of those works and the BuildDepends test passes, and we
> can get MinGW back into working shape, then we can enable ninja for Windows.

I've hoped to get mingw to pass all tests because there is a dependency check,
but it's worth than expected.

>
> -Bill
>
>
>



More information about the cmake-developers mailing list