HDF5 dependency cannot be excluded

[efferre79]

What is the bug?

When configuring the source I can use the cmake define -DGDAL_USE_HDF5=OFF to disable HDF5 support. What happens is that HDF5 is still detected due to the following code in cmake/helpers/CheckDependentLibraries.cmake:

if(HAVE_KEA)
    # CXX is only needed for KEA driver
    gdal_check_package(HDF5 "Enable HDF5" COMPONENTS "C" "CXX" CAN_DISABLE VERSION 1.10)
else()
    gdal_check_package(HDF5 "Enable HDF5" COMPONENTS "C" CAN_DISABLE VERSION 1.10)
endif()

KEA driver depends on HDF5 but if both KEA and HDF5 are disabled, the configure system still looks for HDF5 libraries. I think that something like the following will fix the issue:

if(HAVE_KEA)
    # CXX is only needed for KEA driver
    gdal_check_package(HDF5 "Enable HDF5" COMPONENTS "C" "CXX" CAN_DISABLE VERSION 1.10)
else()
    if(GDAL_USE_HDF5)
        gdal_check_package(HDF5 "Enable HDF5" COMPONENTS "C" CAN_DISABLE VERSION 1.10)
    endif()
endif()

Steps to reproduce the issue

just run cmake with -DGDAL_USE_HDF5=no and -DGDAL_USE_KEA=OFF on a system with hdf5 installed

I get

...

-- Found HDF5: /usr/lib64/libhdf5.so;/usr/lib64/libsz.so;/usr/lib64/libz.so;/usr/lib64/libdl.a;/usr/lib64/libm.so (found version "1.14.4-3") found components: C

...

 * HDF5
   Enable HDF5

...

in the output log.

Versions and provenance

Gentoo Linux

$ gdalinfo --version
GDAL 3.9.1, released 2024/06/22

Additional context

No response

[rouault]

That's probably more a UI / documentation issue than a bug (but not sure where/how clarifying it). Turning off the use of a package doesn't turn off its detection. That's on purpose for reasons I don't remember right now. But the end effect is that drivers that depend on the packages, namely HDF5 and KEA, are not enabled themselves:

-- The following OPTIONAL packages have been found:
[....]
 * HDF5
   Enable HDF5
[....]
-- The following features have been disabled:
[...]
 * gdal_HDF5, Hierarchical Data Format Release 5 (HDF5)
 * gdal_KEA, Kea
[....]

[efferre79]

I see. Thanks for pointing it out. Typically disabling some feature disables also its detection, IMHO it makes sense to cite in the docs. I will leave open there this issue and let you decide if you want to convert into a documentation issue.