whizard @2.2.3_0: update to 2.2.7, use compilers PortGroup

[dstrubbe (David Strubbe)]

Update to version 2.2.7, and use the compilers PortGroup, which will automatically handle the variants for Fortran compilers. The tests are passed for this version on my machine.

[Romendakil]

Actually, make check doesn't work in general on El Capitan as long as SIP is activated as DYLD variables are not exported. make installcheck does work. I don't know how to catch this in the Portfile. From version 2.2.8, WHIZARD will contain C++ code snippets which I have not been testing with clang yet.

[dstrubbe (David Strubbe)]

You can make the test target be 'make installcheck' always -- would that work? At any rate, the issues you are raising are independent of the changes I am suggesting, right?

[Romendakil]

The remarks are both independent, but if you commit your changes into the Macports trunk you might do this change towards 'make installcheck' already now?

[dstrubbe (David Strubbe)]

Patch above committed in r141259.

I tried 'make installcheck' but it fails for me, as below. Some further effort is needed.

:info:test Making installcheck in tests
:info:test PATH=/opt/local/bin:$PATH; export PATH; \
:info:test      LD_LIBRARY_PATH=/opt/local/lib:/opt/local/lib/circe2:$LD_LIBRARY_PATH; \
:info:test      export LD_LIBRARY_PATH; \
:info:test      circe2_tool.opt -test || circe2_tool.bin -test || circe2_tool -test
:info:test /bin/sh: circe2_tool.opt: command not found
:info:test /bin/sh: circe2_tool.bin: command not found
:info:test /bin/sh: circe2_tool: command not found
:info:test make[2]: *** [installcheck-local] Error 127
:info:test make[1]: *** [installcheck-recursive] Error 1
:info:test make: *** [installcheck-recursive] Error 1

[Romendakil]

This is strange, the requested file, circe2_tool.opt, gets installed in ${prefix}/bin so should be available. Are you sure that the CIRCE2 component was correctly built, the requested tool depends on the OCaml compiler which is stated in the depends.

[dstrubbe (David Strubbe)]

The issue is: for MacPorts, 'test' is something that happens before 'install'. So, local copies of such files need to be used rather than their locations where they would be installed.

[Romendakil]

Understood. Then maybe we should just disable the 'check' option. I know this is not good practice either but the danger is that someone will have El Capitan with SIP on, and then the testsuite were to fail and, hence, installation, too.

[dstrubbe (David Strubbe)]

How about this? Also, keep in mind that the test phase is not run unless the user specifically requests it via "port test". If they just do "port install", no tests are run. So, test failures will not block installation.

-test.run            yes
-test.target         check
+# make check doesn't work in general on El Capitan as long as SIP is activated as DYLD variables are not exported.
+# make installcheck works, but needs patches to work from the build directory rather than installed locations
+if {${os.major} > 11} {
+     test.run            yes
+     test.target         check
+ }

[Romendakil]

Ah, ok, I didn't know that 'make check' is done only when you invoke 'port test'. This is interesting. No, then your solution seems a good one. Shouldn't it be 'if {$os.major} < 11} { .....} however?

[dstrubbe (David Strubbe)]

Oh, yes of course. Also I realize the number wasn't correct (​http://trac.macports.org/wiki/PortfileRecipes#compare-osx-darwin-version). That part committed in r141303.

I tried making an option to do installcheck as part of the destroot phase (commented out in the commit). It works in the circe2 and omega parts (with a patch), though fails for functional_tests: probably another patch is needed there.

:info:destroot ============================================================================
:info:destroot Testsuite summary for WHIZARD 2.2.7
:info:destroot ============================================================================
:info:destroot # TOTAL: 187
:info:destroot # PASS:  1
:info:destroot # SKIP:  20
:info:destroot # XFAIL: 1
:info:destroot # FAIL:  163
:info:destroot # XPASS: 2
:info:destroot # ERROR: 0
:info:destroot ============================================================================
:info:destroot See tests/functional_tests/test-suite.log
:info:destroot ============================================================================

[Romendakil]

Can you see from the log files what is going wrong?

[dstrubbe (David Strubbe)]

No, it doesn't say anything more, and it's not clear where (if anywhere) the test-suite.log file ended up. I would guess it is similar issues to what I patched.

[Romendakil]

Ok, then let it be, and we do the fix you proposed by excluding the checks for anything as new as El Capitan.