cmake: fix a few constraints in dependents#3259
Conversation
cc80851 to
671997b
Compare
|
Now testing with the fix included from spack/spack#51923. Edit: seems to work, reverting that... |
76fa0b6 to
7691c31
Compare
* use `type=build` where appropriate
* avoid a "global" `conflicts("cmake`, use `depends_on("cmake@...`
instead
Signed-off-by: Harmen Stoppels <me@harmenstoppels.nl>
2c838b8 to
545665b
Compare
|
Rebased this to include the latest Spack version to deal with the concretization issue... |
Signed-off-by: Harmen Stoppels <me@harmenstoppels.nl>
|
@johnwparent for your information: the last commit removes your constraint that requires
Maybe you could bump external CMake to a newer version in the relevant image and add back the constraint? IMO the change shouldn't be controversial given that the conflict was ignored from the start and paraview built fine. |
|
There's also always the potential that updating the image might cause other CI failures, so if this is an immediate need, it makes sense to just merge without trying to update the CI image. I don't love actively introducing a broken relationship between CMake and We also must not actually be using the fortran compiler with CMake in CI, since otherwise I have no idea how that's working... maybe there's a lower bound where oneAPI + MSVC + CMake work again... The correct solution, as Harmen points out, is to bump the CMake version in the CI image to post 4.1 (and for me to investigate whether there is a lower bound for this conflict) but that may take some time if the image updates introduces other failures. Thanks for the heads up @haampie |
This reverts commit c89894a.
Signed-off-by: Harmen Stoppels <me@harmenstoppels.nl>
|
Just commented out the constraint so that we don't forget about it |
Fixes a couple oversights that may or may not prevent unified concretization of
build dependencies in the Windows stack.