Targets for "make test" seem somewhat truncated in Makefile

[GoogleCodeExporter]

What steps will reproduce the problem?
1. try "make test".
2. see it fail.
3. look in Makefile around line 95-97.

What is the expected output? What do you see instead?
The tests can't be launched.
I see $(UNITTESTARGS) as the beginning of the command line (an argument, no 
command).
UNITTESTARGS seems used without being defined.
Hence bad command line.

What version of the product are you using? On what operating system?
0.5.0+svn on Linux.

Please provide any additional information below.
Other targets seem to suffer the same issue. See test-with-cover.

Context: before trying to write a db_hanoidb disk backend driver, I'd like to 
see tests running for original scalaris ; then run them against db_hanoidb.

Original issue reported on code.google.com by adrp...@gmail.com on 5 Feb 2013 at 1:17

[GoogleCodeExporter]

Please rerun configure. I guess that you get the following error message:
"neither run_test nor ct_run found - on erlang < R14 consider running 
install.sh in the common_test directory otherwise you won't be able to run the 
unit tests"
You need either run_test or ct_run to run the unit-tests. It depends on your 
Linux distribution, but in my case ct_run is part of the erlang-common_test 
package.

Original comment by schu...@gmail.com on 5 Feb 2013 at 1:49

[GoogleCodeExporter]

Yes, you narrowed the issue, thanks for your fast response.
I run R15B1. ls /usr/lib/erlang/bin/ gives me ct_run, not run_test.

I made
make clean
then
./configure
and I see

(snip)
checking for Erlang/OTP 'xmerl' library subdirectory... 
/usr/lib/erlang/lib/xmerl-1.3.1
checking for Erlang/OTP 'xmerl' library version... 1.3.1
checking for Erlang/OTP 'common_test' library subdirectory... 
/usr/lib/erlang/lib/common_test-1.6.1
checking for Erlang/OTP 'common_test' library version... 1.6.1
checking for Erlang/OTP 'edoc' library subdirectory... 
/usr/lib/erlang/lib/edoc-0.7.9.1
checking for Erlang/OTP 'edoc' library version... 0.7.9.1
checking for /usr/lib/erlang/lib/common_test-1.6.1/priv/bin/run_test... no
checking for /usr/lib/erlang/bin/run_test... no
checking for run_test... no
configure: run_test not found - on erlang < R14 consider running install.sh in 
the common_test directory otherwise you won't be able to run the unit tests
checking for Erlang/OTP 'toke' library subdirectory... not found
checking for Erlang/OTP 'toke' library version... not found
configure: toke erlang library not found, disabling toke support...
(snip)

So I have R15 and a working Common Test (I have run it ok on another project).
I thought the issue was with the Makefile. It seems rather related to 
configure.ac around lines 151~158. Is it supposed to work with R15?

Pierre M. in need of one more clue

Original comment by adrp...@gmail.com on 5 Feb 2013 at 5:35

[GoogleCodeExporter]

Do you have by any chance an old version of configure? Before January 21st, 
configure would only check for run_test and give up if there is no such file. 
With r4349, we added an additional test for ct_run.

You mentioned that the checks are around lines 151~158. But in the most recent 
version, they are on lines 151~167.

Please run:
svn up configure configure.ac
./configure
This should hopefully fix your problem.

Original comment by schu...@gmail.com on 5 Feb 2013 at 7:14

[GoogleCodeExporter]

Hello,

yes, you are right:
svn up configure configure.ac
gave me revision 4466 of those files and now after ./configure the target of 
"make test" has a nice /usr/lib/erlang/bin/ct_run as a command. Yeah ! Thanks.

Unfortunately after "make clean" and "make" (which ran ok) "make test" still 
has some issue :
/usr/lib/erlang/bin/ct_run  -name ct  -pa `pwd`/ebin `pwd`/test 
`pwd`/contrib/yaws/ebin `pwd`/contrib/log4erl/ebin -spec 
test/scalaris.runtests-default.cfg -ct_hooks scalaris_cth -noshell | tee ct.log 
; \
    grep " 0 failed" ct.log
{error_logger,{{2013,2,6},{11,7,32}},"Can't set long node name!\nPlease check 
your configuration\n",[]}
{error_logger,{{2013,2,6},{11,7,32}},crash_report,snip...

Aren't those `pwd` funny ?-)
I did a whole svn update (I'm now with 4466 scalaris wide, not just configure) 
and I tried:
make clean (ok)
make (ok)
time /usr/lib/erlang/bin/ct_run -name ct -pa ./ebin ./test ./contrib/yaws/ebin 
./contrib/log4erl/ebin -spec test/scalaris.runtests-default.cfg -ct_hooks 
scalaris_cth -noshell | tee ct.log ; \
 grep " 0 failed" ct.log

but I got the same result:
{error_logger,{{2013,2,6},{11,17,46}},"Can't set long node name!\nPlease check 
your configuration\n",[]}
{error_logger,{{2013,2,6},{11,17,46}},crash_report,snip...

In test/scalaris.runtests-default.cfg I don't see anything about node names, 
nothing about length of node names.

I newbiely think I still have an issue with "configure" (because of the `pwd` 
in the Makefile) and I'm lost with the "long node name" error message.
Thank you for your assistance so far. I hope you have another clue to help me 
go forward on "make test".

Pierre M.

Original comment by adrp...@gmail.com on 6 Feb 2013 at 10:28

[GoogleCodeExporter]

Does bin/scalarisctl checkinstallation run without errors or gives you some 
hints what the problem may be?

Does 'hostname -f' on your machine result in a full qualified domain name 
(foo.bar.com) and not only in a hostname (foo)?

Original comment by schin...@gmail.com on 6 Feb 2013 at 1:30

[GoogleCodeExporter]

Seems you are right again (hostname issue?)

bin/scalarisctl checkinstallation
outputs:
Running basic tests...
INFO: could not find Scalaris' Java-API files. You won't be able to use the 
'scalaris' command line script to access Scalaris.
  'make java' will build the Java-API
We were trying to run: 
/home/(snip)/DBMS/scalaris/scalaris-read-only/java-api/scalaris --noconfig -h
Running Scalaris run-time tests...
  NOTE: yaws port not specified.
        Python, Python3, Ruby API tests may fail if the port is different in
        one of the config files.
  Java-API ... NOT INSTALLED
  Python-API ... SUCCESS
  Python3-API ... NOT INSTALLED
  Ruby-API ... SUCCESS

(I had not "make java" or anything else, only "make". I only do erlang 
scalaris).

hostname -f give only one word(foo), not a F.Q.D.N (foo.bar.org). There is 
something localy wrong here on my (Debian Testing) system. But it doesn't 
prevent me to run several nodes (n1 to n5)@127.0.0.1 to setup a scalaris 
cluster. Is there a similar way to give "localhosty" names for ct_run to launch 
ok?

Original comment by adrp...@gmail.com on 6 Feb 2013 at 4:26

[GoogleCodeExporter]

Can you try the following?
SCALARIS_CT_NAME=ct@127.0.0.1 make test
Maybe it works around your hostname issues.

Original comment by schu...@gmail.com on 6 Feb 2013 at 10:02

[GoogleCodeExporter]

Yes, thanks,
SCALARIS_CT_NAME=ct@127.0.0.1 make test
makes things better. ct_run has run for more than 15 minutes and I got HTML 
reports.
Here is the end of the console output:
Testing scalaris.scalaris-read-only: TEST COMPLETE, 80 ok, 26 failed, 383 
skipped of 489 test cases

Updating /home/(snip)/DBMS/scalaris/scalaris-read-only/index.html... done
Updating /home/(snip)/DBMS/scalaris/scalaris-read-only/all_runs.html... done

make: *** [test] Erreur 1

I often see such things in the console:
2013-02-07 14:16:12.468
Trying to build Scalaris with ids

=ERROR REPORT==== 7-Feb-2013::14:16:12 ===
Error in process <0.3610.0> on node 'ct@127.0.0.1' with exit value: 
{badarg,[{gen_component,start,5,[{file,"src/gen_component.erl"},{line,373}]}]}

In index.html I see:
Ok=80
Failed=26
Skipped(User/Auto)=383(10/373)
Missing Suites=0
Node=ct@127 (NOT ct@127.0.0.1)

In suite.log.html
for SKIPPED (orange) lines
I see "init_per_suite failed" or "init_per_testcase timed out"  or 
"init_per_testcase failed" or "init_per_suite timed out" comments (or "ignore 
benchmarks" or some other messages).
The last line of that big table is:
TOTAL 1174.197s FAILED 80 Ok, 26 Failed, 10 Skipped of 116

How can we diagnose those failed/timeout inits to have less tests skipped?

Original comment by adrp...@gmail.com on 7 Feb 2013 at 1:44

[GoogleCodeExporter]

SCALARIS_CT_NAME=ct@127.0.1.1 make test
gave same result (strange Debian 0.1.1 thing)

Original comment by adrp...@gmail.com on 7 Feb 2013 at 2:15

[GoogleCodeExporter]

Note: Skipped tests are fine they were deactivated by us.

Original comment by schin...@gmail.com on 7 Feb 2013 at 3:28

[GoogleCodeExporter]

Every time you run the unit-tests Erlang will create a directory for the log- 
and html-files. The directory's name is something like 
ct_run.ct@127.0.0.1.2013-.... Could you send me a copy of this directory (if 
you click on schuett, you can get my email address)? We are wondering which 
tests fail and whether we can see a pattern.

Does your loopback-device support 127.0.0.1 as well as 127.0.1.1? Or is it only 
one of them? Unfortunately, I am not that familiar with Debian.

Original comment by schu...@gmail.com on 7 Feb 2013 at 3:35

[GoogleCodeExporter]

Hello again,
I have made clean, svn update, rerun configure and make. ct@127.0.0.1 seems 
incorporated now and "make test" has same numerical results (80 Ok, 26 Failed, 
10 Skipped of 116).
But good news, read on.

In ct.log I also see:
2013-02-07 15:35:04.545
Starting unittest [{current,{api_json_SUITE,init_per_suite}},
                   {successful,0},
                   {failed,0},
                   {skipped,{0,0}},
                   {total,0}]

----------------------------------------------------
2013-02-07 15:35:04.563
unittest_helper:make_ring size 4

----------------------------------------------------
2013-02-07 15:35:04.665
[error] [ CC ] can't listen on 14195: eaddrinuse

----------------------------------------------------
2013-02-07 15:35:04.666
[error] Error: exception error:function_clause in comm_acceptor:init/2:  
[{prim_inet,sockname,[abort],[]},
 {comm_acceptor,init,2,[{file,"src/comm_layer/comm_acceptor.erl"},{line,51}]}]

----------------------------------------------------
2013-02-07 15:35:04.666
[error] Error in process <0.1329.0> on node 'ct@127.0.0.1' with exit value: 
{function_clause,[{comm_acceptor,init,2,[{file,"src/comm_layer/comm_acceptor.erl
"},{line,62}]}]}

I investigated this eaddrinuse with netstat.
There wera losts of old BEAM process there.
killall epmd and killall beam.smp helped to clean.
After this pretesting cleanup, "make test" was much better:
Testing scalaris.scalaris-read-only: TEST COMPLETE, 537 ok, 0 failed, 10 
skipped of 547 test cases

Yeah! Thanks for the working update.
And congrats to all test writers: nice job.

Little suggestion by the way:
ERL_EPMD_ADDRESS=127.0.0.1
if it makes sense.

Pierre M.

Original comment by adrp...@gmail.com on 7 Feb 2013 at 3:35

[GoogleCodeExporter]

Hello,

why is this issue not closed? Thanks to your help, it now "works for me".

Pierre M.

Original comment by adrp...@gmail.com on 13 Feb 2013 at 1:37

[GoogleCodeExporter]

you are right - now with 537 ok, 0 failed, 10 skipped, everything seems fine

Original comment by nico.kru...@googlemail.com on 13 Feb 2013 at 1:42

• Changed state: Fixed