fix cdf5 configure option

[wkliao]

This PR fixes configure/cmake that fails to disable CDF5 on 64-bit machines when user explicitly disables CDF5. It also errors out when user explicitly enable CDF5 on 32-bit machines (size_t being of size 4 bytes).

[edhartnett]

Change to "whether CDF5 support should be disabled"

[DennisHeimbigner]

If you get rid of the line

-OPTION(ENABLE_CDF5 "Enable CDF5 Support." ON)
isn't there a cmake command to list the available options (like configure --help)? This would mean
that ENABLE_CDF5 will not show up as an explicit option?

[wkliao]

If we kept that line, then ENABLE_CDF5 will be set to either ON or OFF. Unlike autoconf which allows us to set to a customized value, AUTO in this case, meaning user did not explicitly set this option.

I am not aware cmake has a similar command-line option to --help that shows all available options. If you know a way to do so, I am interested to learn.

[DennisHeimbigner]

cmake -L
althought it is a bit noisy.

[wkliao]

Do people actually run that command to check all available options?
I don't know what to add to achieve the same effect as OPTION, if we want to know whether user explicitly enable or disable an option.

[DennisHeimbigner]

I have used it but rarely. I suspect however, that people do peruse
a CMakeLists.txt looking for option() commands.

[DennisHeimbigner]

Ok, I think understand. I would only suggest
that you leave the option() command at line
at line 1290, but commented out with a short
explanation of the fact that the option command does not allow a way to test if the user actually
set the flag or not.

[wkliao]

This commit should achieve the same effect of using OPTION.
The trick learned from
https://cmake.org/pipermail/cmake/2016-October/064342.html

[DennisHeimbigner]

That seems like a good solution. Wonder why the cmake developers
did not support tristate from the get-go. Oh well.

[WardF]

It doesn't matter so we will accept this, but as a point of reference cmake will support 'ON/OFF' and 'TRUE/FALSE' interchangeably.

[wkliao]

I thought it was interchangeable, but when I used -DENABLE_CDF5=FALSE, it does not disable CDF5. Adding a MESSAGE command to print out the value of ENABLE_CDF5 and I got its value ON. Please give it a try. I hope it is not just me. (Note this applies only to the master branch before my PR was merged.)

[wkliao]

In addition, my PR changed the type of ENABLE_CDF5 from BOOL to STRING. So using TRUE/FALSE will fail.

[WardF]

Using -DENABLE_[WHATEVER OPTION]=FALSE disables the corresponding feature on my system. I just tested -DENABLE_NETCDF_4=FALSE, and saw netcdf4 API support turn off. I saw similar behavior when testing -DENABLE_CDF5=FALSE, before accepting the pull request; I tested before making my comment.

[WardF]

We can see from the cmake documentation found at https://cmake.org/cmake/help/v3.12/command/if.html that OFF and FALSE are treated as the same. We will need to update the new logic in CMakeLists.txt so that it respects valid CMake boolean specifiers as users would expect, before the next release.

[WardF]

I am seeing test failures on ARM that exist with this merge. I am going to revert this pull request for now as I'm about to go out of town and won't be available to debug this. We can reevaluate this pull request after I return, and figure out what is going on.

Failures can be seen here:

• http://cdash.unidata.ucar.edu/viewTest.php?onlyfailed&buildid=3790