Fixing build on non-ib systems

[shamisp]

RC/UD build was enabled out of the scope of IB build

[shamisp]

@yosefe please review

[mellanox-github]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
http://bgate.mellanox.com:8000/jenkins-secure/job/gh-ucx-pr/37/

Build Log
last 50 lines

[...truncated 805 lines...]
  CC       stats/libstats.lo
  CC       stats/stats_parser.o
  CXX      gtest/libucstest_la-test_helpers.lo
  CXX      gtest/libucstest_la-main.lo
  CXX      gtest/libucstest_la-test.lo
  CCLD     libucs.la
  CCLD     ucs_stats_parser
  CXXLD    libucstest.la
make[2]: Leaving directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/src/ucs'
Making all in src/uct
make[2]: Entering directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/src/uct'
  CC       tl/libuct_la-context.lo
  CC       ib/base/libuct_la-ib_context.lo
  CC       ib/base/libuct_la-ib_device.lo
  CC       ib/base/libuct_la-ib_iface.lo
  CC       ib/rc/libuct_la-rc_iface.lo
  CC       ib/rc/libuct_la-rc_ep.lo
  CC       ib/mlx5/libuct_la-ib_mlx5.lo
  CCLD     libuct.la
make[2]: Leaving directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/src/uct'
Making all in test/perf
make[2]: Entering directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/test/perf'
  CC       libperf.lo
  CC       perftest.o
  CCLD     libucxperf.la
  CCLD     ucx_perftest
/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/src/uct/.libs/libuct.so: undefined reference to `uct_rc_mlx5_tl_ops'
collect2: ld returned 1 exit status
make[2]: *** [ucx_perftest] Error 1
make[2]: Leaving directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/test/perf'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test'
make: *** [all] Error 2
Build step 'Execute shell' marked build as failure
TAP Reports Processing: START
Looking for TAP results report in workspace using pattern: **/*.tap
Did not find any matching files.
[Valgrind] Files to copy:
[Valgrind] Analysing valgrind results
[Valgrind] Ending the valgrind analysis.
Anchor chain: could not read file with links: /var/lib/jenkins/jobs/gh-ucx-pr/workspace/jenkins_sidelinks.txt (No such file or directory)
[copy-to-slave] The build is taking place on the master node, no copy back to the master will take place.
Email was triggered for: Failure - Any
Sending email for trigger: Failure - Any
Request made to compress build log
Sending email to: eugene@mellanox.com miked@mellanox.com yosefe@mellanox.com
[BFA] Scanning build for known causes...

[BFA] Done. 0s

Test FAILed.

[shamisp]

@yosefe - can you please take a look what happened on MLNX machine ?

[mike-dubman]

@shamisp - you can click on link (see above "Details" and go to "Console log"
http://bgate.mellanox.com:8000/jenkins-secure/job/gh-ucx-pr/37/console
Also, we can provide you with ssh access to the machine (I have you have one already with your ssh key)

[mike-dubman]

@shamisp , @yosefe - btw, maybe it makes sense to add "--without-ib" test into jenkins to mimic this behave.

[yosefe]

@shamisp Your fix is somewhat surprising; if IB is not present, then, mlx5_hw.h header would not be present anyway. Also, you moved it from the scope of IB to the scope of RC..

[yosefe]

RC transport would not be enabled unless IB is enabled.. if It is, it would be considered a bug in ib.m4 script

[shamisp]

I think now it is automatically detected (which is good) but the detection is broken. So this is something that I want to solve.
For regular case I don't want to explicitly specify that I don't want ib.=

[shamisp]

This is the bug in M4. RC enables IB when it is not supposed to do so.

[shamisp]

The same true for dc,ud,mlx5

[mellanox-github]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
http://bgate.mellanox.com:8000/jenkins-secure/job/gh-ucx-pr/41/

Build Log
last 50 lines

[...truncated 1009 lines...]
[       OK ] test_datatype.notifier_chain (0 ms)
[----------] 11 tests from test_datatype (1110 ms total)

[----------] 6 tests from test_config
[ RUN      ] test_config.parse_default
[       OK ] test_config.parse_default (0 ms)
[ RUN      ] test_config.parse_with_prefix
[       OK ] test_config.parse_with_prefix (0 ms)
[ RUN      ] test_config.clone
[       OK ] test_config.clone (0 ms)
[ RUN      ] test_config.set
[       OK ] test_config.set (0 ms)
[ RUN      ] test_config.performance
[       OK ] test_config.performance (2 ms)
[ RUN      ] test_config.dump
[       OK ] test_config.dump (0 ms)
[----------] 6 tests from test_config (2 ms total)

[----------] 1 test from test_component
[ RUN      ] test_component.init_cleanup
[       OK ] test_component.init_cleanup (0 ms)
[----------] 1 test from test_component (0 ms total)

[----------] 1 test from test_uct_perf
[ RUN      ] test_uct_perf.envelope
/scrap/jenkins/jobs/gh-ucx-pr/workspace/contrib/../test/gtest/uct/test_uct_perf.cc:183: Failure
Error: No such device
terminate called after throwing an instance of 'ucs::test_abort_exception'
  what():  std::exception
/scrap/jenkins/jobs/gh-ucx-pr/workspace/contrib/../test/gtest/uct/test_uct_perf.cc:183: Failure
Error: No such device
make: *** [test] Aborted
make: Leaving directory `/scrap/jenkins/jobs/gh-ucx-pr/workspace/build-test/test/gtest'
Build step 'Execute shell' marked build as failure
TAP Reports Processing: START
Looking for TAP results report in workspace using pattern: **/*.tap
Did not find any matching files.
[Valgrind] Files to copy:
[Valgrind] Analysing valgrind results
[Valgrind] Ending the valgrind analysis.
Anchor chain: could not read file with links: /var/lib/jenkins/jobs/gh-ucx-pr/workspace/jenkins_sidelinks.txt (No such file or directory)
[copy-to-slave] The build is taking place on the master node, no copy back to the master will take place.
Email was triggered for: Failure - Any
Sending email for trigger: Failure - Any
Request made to compress build log
Sending email to: eugene@mellanox.com miked@mellanox.com yosefe@mellanox.com
[BFA] Scanning build for known causes...

[BFA] Done. 0s

Test FAILed.

[shamisp]

@miked-mellanox - I think my fix is ok now, but the test fails silence no transport is available it is trying to use hardcoded transport. (mlx5_0). In the #24 you solving similar issue. Feel free to merge it my patch since it seems to be function. Let me know how do we want to proceed.

[mike-dubman]

@shamisp , it is fixed in #22 (the m4/ib )

[mike-dubman]

i think that detection logic in #22 is better and covers more cases.

[shamisp]

@miked-mellanox #24 (Mellanox) or #22(ORNL) ? Which one do you prefer ?
Let's merge in one of the branches. I tested both and both seems to be functional on my Fedora20.

[shamisp]

solved in #24