[cmake-developers] Java language support regression (compared with 3.7.x and 3.8.x) for 3.9.0

[Alan W. Irwin]

I have built CMake-3.9.0 on my Linux (Debian Jessie) platform using
the bootstrap method I have always successfully used for other CMake
versions (including 3.7.2).

For a complex project (PLplot) which uses swig_add_module and
swig_link_libraries to implement our Java binding, I get the following
error message at the end of that CMake-3.9.0 output.

-- Configuring done
CMake Error: Error required internal CMake variable not set, cmake may
not be built correctly.
Missing variable is:
CMAKE_Java_CREATE_SHARED_MODULE
-- Generating done
-- Build files have been written to:
/home/software/plplot/HEAD/build_dir

I don't get this error message for CMake-3.7.2.

Using find and xargs -0 grep I have looked in both 3.7.2 and 3.9.0
source code for any mention of CMAKE_Java_CREATE_SHARED_MODULE, and it
just does not exist.

By the way, I have tried the following simple project.

cmake_minimum_required(VERSION 3.6.2 FATAL_ERROR)
project(test_java NONE)
enable_language(Java)
message(STATUS "CMAKE_Java_CREATE_SHARED_MODULE = ${CMAKE_Java_CREATE_SHARED_MODULE}")

(I used 3.6.2 for the minimum version because that is what PLplot uses).

That project does confirm that both 3.7.2 and 3.9.0 do not define
CMAKE_Java_CREATE_SHARED_MODULE.  However, both versions failed to
generate the above error so a more complex project such as PLplot
that, e.g., swig-generates a Java binding for a C library is needed to
show the issue.

By the way, inserting the following lines in the PLplot top-level
CMakeLists.txt file

# Temporary workaround for Java language support issue for
# CMake-3.9.0
set(CMAKE_Java_CREATE_SHARED_MODULE "this_command_should_never_be_used")

has no bad consequences for cmake-3.7.2, i.e., I can build our Java
binding and run all our standard Java examples.  However, for the
3.9.0 case, the above change avoids the cmake issue, but there are
many JNI-related build errors for our Java binding (although none
of them mention "this_command_should_never_be_used").

The only cmake-3.8.0 version I have built previously is one of the
release candidates.  But that passed all tests at the time including
building our Java binding and running our Java examples with success
so this appears to be a newly introduced issue for 3.9.0.

>From these symptoms, I am pretty sure that language support infrastructure
has been upgraded for 3.9.0, but your Java
language support has not been consistently upgraded at the same time.

If so, the solution is either downgrade your language support
infrastructure to be consistent with CMake-3.8.x or to upgrade your
Java language support to be consistent with your updated language
support for 3.9.0.

Once someone has a patch for this issue I would be happy to test it by
rebuilding a patched version of CMake-3.9.0 and trying out a build of
PLplot with that patched version.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state
implementation for stellar interiors (freeeos.sf.net); the Time
Ephemerides project (timeephem.sf.net); PLplot scientific plotting
software package (plplot.sf.net); the libLASi project
(unifont.org/lasi); the Loads of Linux Links project (loll.sf.net);
and the Linux Brochure Project (lbproject.sf.net).
__________________________

Linux-powered Science
__________________________