Replace resource expensive bcrypt test with shorter version

[eiri]

Since we are using 3rd party app for bcrypt and couch_passwords: bcrypt/2 is just an interface for it there are no need to test correctness of the lib itself, just the fact that we are using it properly.

Test bcrypt_logRounds_18 is kind of a resource hog, so I suggest to remove it to easy a burden on Travis and Jenkins. This reduces the test runtime on my laptop from 75s to 15s and I suspect the difference for CI will be even greater.

[eiri]

@wohali @janl @rnewson Any opinion on this?

[iilyak]

good idea.

[wohali]

We don't need to validate that the algorithm works when logrounds has been increased - just that it's able to function correctly when linked and used.

Good call.

[wohali]

Oh, thought - maybe make it bcrypt_logRounds_5 instead, or some other (lower) number, as a way to validate that you can adjust the figure and the library still does the right thing?

[eiri]

@wohali I've added bcrypt_logRounds_5 and (surprise!) found that the tests were actually broken - generators been returning test objects instead of running assertions, infamous and hugely common error.

I fixed that, simplified setup (we've been starting and stopping bunch of unused apps and for some reason did it for each test individually) and added the tests descriptions. How does it look for you now?

[wohali]

Good catch! +1 and go for it.

[wohali]

@eiri we got a random failure in Jenkins related to encryption after this hit master, can you investigate please?

https://builds.apache.org/blue/organizations/jenkins/CouchDB/detail/master/223/pipeline

Logs at https://couchdb-vm2.apache.org/ci_errorlogs/jenkins-couchdb-223-2018-03-23T23:15:34.663810/couchlog.tar.gz

test/javascript/tests/users_db_security.js                     
    Error: undefined
Trace back (most recent call first):
    
  52: test/javascript/test_setup.js
      T((void 0))
  35: test/javascript/tests/users_db_security.js
      ("jchris1")
  39: test/javascript/tests/users_db_security.js
      ([object Object],"org.couchdb.user:fdmanana","jchris1")
 168: test/javascript/tests/users_db_security.js
      ("pbkdf2",(function (derived_key) {TEquals(40, derived_key.length, "de
 413: test/javascript/tests/users_db_security.js
      ()
 395: test/javascript/couch_test_runner.js
      run_on_modified_server([object Array],(function () {try {testFun(schem
 411: test/javascript/tests/users_db_security.js
      ("pbkdf2",0,[object Array])
 399: test/javascript/tests/users_db_security.js
      ()
  37: test/javascript/cli_runner.js
      runTest()
  48: test/javascript/cli_runner.js

[eiri]

@wohali It's hard to imagine that fixing eunit tests would break API tests.

I looked into it and as far as I can see with bcrypt PR API test users_db_security.js was changed to be prune to racing condition - user database got deleted and re-created, which is known anti-pattern.

[wohali]

Thanks @eiri, sorry for leaving this request here. I wasn't accusing you of making the mistake :) I'll open a new ticket. Thank you again!