Re: math/blas build failure - cmake lossage

["David H. Gutteridge" <david%gutteridge.ca@localhost>]

On Wed, 20 Sep 2023, at 20:40:45 -0400, Greg Troxel wrote:
>math/blas fails on netbsd-10 amd64 with pkgsrc up to date:
>
>  -- Build html documentation: OFF
>  -- Build man documentation: OFF
>  -- Configuring done (0.6s)
>  CMake Error: CMake can not determine linker language for target:
run_test_zcomplexabs
>  CMake Error: CMake can not determine linker language for target:
run_test_zcomplexdiv
>  CMake Error: CMake can not determine linker language for target:
run_test_zcomplexmult
>  CMake Error: CMake can not determine linker language for target:
run_test_zminMax
>  -- Generating done (0.1s)
>  CMake Generate step failed.  Build files cannot be regenerated
correctly.
>  *** Error code 1
>
>Digging into CMakeLists.txt, I find:
>
>where run_test_zcomplexabs has both a command and sources.
>
>
>  add_executable( test_zcomplexabs
${LAPACK_SOURCE_DIR}/INSTALL/test_zcomplexabs.f )
>  add_custom_target( run_test_zcomplexabs
>    COMMAND test_zcomplexabs 2> test_zcomplexabs.err
>    WORKING_DIRECTORY ${LAPACK_BINARY_DIR}/INSTALL
>    COMMENT "Running test_zcomplexabs in ${LAPACK_BINARY_DIR}/INSTALL
with stderr: test_zcomplexabs.err"
>    SOURCES ${LAPACK_SOURCE_DIR}/INSTALL/test_zcomplexabs.f )
>
>
>dropping SOURCES makes it build.
>
>
>Does anybody understand cmake enough to know what is right?

This seems to be a regression with CMake 3.27.5, which pkgsrc was
updated to on September 16th. It has evidently been fixed in 3.27.6:
https://gitlab.kitware.com/cmake/cmake/-/commit/978b68d3bbb44e3bc6adbe5109eb8295f8232ce6
(I haven't tested this.) So our fix would be to revert to .4 or go
forward to .6.

Dave