improve parallel builds, fixed SEGFAULT for NULL parameter for nc_inq_format(), added testing, documentation

[edhartnett]

This PR started with adding some testing for ncfunc.c, which revealed a SEGFAULT when a NULL parameter is passed to nc_inq_format(). This is fixed in this PR.

Also I continued work on autotools build system for smoother parallel builds. Many test dependencies are eliminated by ensuring that all tests use unique file names for test output. Parallel builds now work for builds with dap remote tests, and buildfiles are simplified by removal of unneeded dependencies.

Also took out some dead code and added doxygen documentation for the functions I touched.

Fixes #678.
Fixes #677.
Fixes #682.
Fixes #681.
Fixes #670.
Fixes #616.
Fixes #667.
Fixes #687.
Fixes #685.
Fixes #689.
Fixes #688.
Fixes #684.
Fixes #690.
Fixes #695.
Fixes #693.

[WardF]

Why did you remove tst_filters? That's part of the new code added by Dennis for multiple compression filters, I believe. @DennisHeimbigner is that correct?

[edhartnett]

Sorry, didn't remove tst_filters, remove tst_formats. Just mistyped it in git commit comment. ;-)

I added tst_formats.c and am having some trouble getting it to work with travis. I also added a ref_.hdf4 file, and that seems to be part of the trouble. I will keep working it until I figure it out...

[DennisHeimbigner]

The file test_filters is part of the filter_test directory.
There is a specific ./configure flag to control if the tests
in filter_test are attempted: --enable-filter-tests
This defaults to off because this test is a bit tricky to run
automatically and as far as I recall, I was never able to get it to
run under cmake.
It looks like the equivalent option is not defined in CMakeLists.txt.
It should be and it should default to off. The relevant place it hould be used
is in nc_test4/filter_test/CMakeLists.txt.
See how it is used in nc_test4/filter_test/Makefile.am.

[edhartnett]

OK, this is now working for both autotools and cmake builds.

As noted above, I did NOT remove tst_filters, just mistyped "tst_formats" in a commit comment. No tests were harmed in the production of this pull request.

The new tst_foramats is added to test nc_inq_foramt() and nc_inq_format_extended().

[WardF]

Evaluating now, thanks!

[WardF]

CMake configuration breaks using my automated scripts on Windows, but probably something basic. I'll investigate.

[edhartnett]

OK, keep me in the loop. I am happy to revert some files to eliminate the problem and get the rest of the changes through, then I can circle back around to the problem area.

Unfortunately windows is the one giant gap in my CI. I don't have a windows machine with visual studio. Not sure if there is a free version I can use or whether I need to buy one...

[WardF]

The issue is only on Windows, so I'm sure it's pretty straightforward; I'll outline what I find and if it's something I can't fix myself. First avenue of investigation is my automated tools to see if they're doing something they shouldn't, now.

[DennisHeimbigner]

Ed, FYI the community edition of Visual Studio is free.

[edhartnett]

I have found an old windows laptop. I will fire it up and see what I can manage wrt visual studio community edition. ;-)

[WardF]

No prob I had to set this aside suddenly yesterday (son sick at school) but sitting down to it now:)

[WardF]

I found the issue, it's no big deal, when MSVC is true a property is being set that doesn't need to be set. This is from some older cmake code that could do with cleaning up; cmake has come a long way since we started using it, plus I can refine a lot of what's been done. I'll take care of it and get this merged, thanks!

[WardF]

Also FWIW I do my debugging/testing for Windows in a Win10 VM using VirtualBox, with no ill effects.

[edhartnett]

Well the problem with virtual machines is: when you get mad, what do you kick? ;-)

…

On Wed, Nov 29, 2017 at 12:17 PM, Ward Fisher ***@***.***> wrote: Also FWIW I do my debugging/testing for Windows in a Win10 VM using VirtualBox, with no ill effects. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#680 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEUr3Oi3mCWCBMcZvLSqAc3pYQx_uQFRks5s7a3BgaJpZM4QnvDn> .

[edhartnett]

@WardF hope your son is feeling better. ;-(

Meanwhile I have been working on PIO issues all day so still have not yet set up a test environment...

[WardF]

Running mpich parallel tests and then will merge. @edhartnett one thing our (docker-based) regression testing does is it tests the Fortran, C++ and Python interfaces, as well as NCO (for non parallel builds) and parallel-netcdf (for parallel builds) against the particular C library being generated. I don't know if that's something your Jenkins build is set up to do, and to be honestly it only rarely finds an issue, but it is pretty handy when it does.

Thanks!