[metrics-bugs] #22733 [Metrics/Library]: Use parameterized tests instead of repeated methods

Tor Bug Tracker & Wiki blackhole at torproject.org
Mon Oct 16 14:40:49 UTC 2017


#22733: Use parameterized tests instead of repeated methods
-----------------------------+------------------------------
 Reporter:  iwakeh           |          Owner:  metrics-team
     Type:  enhancement      |         Status:  new
 Priority:  Medium           |      Milestone:
Component:  Metrics/Library  |        Version:
 Severity:  Normal           |     Resolution:
 Keywords:  metrics-2018     |  Actual Points:
Parent ID:                   |         Points:
 Reviewer:                   |        Sponsor:
-----------------------------+------------------------------

Comment (by karsten):

 I mostly agree that parameterized tests make the code more readable.
 However, there's one thing that makes them a bit less readable: without a
 test method name, whoever reads the code needs to guess why given test
 input produces given test output. That's not always obvious just by
 looking at the data.

 As a (truly) random example, look at the various `testSearchBase64*`
 methods
 [https://gitweb.torproject.org/onionoo.git/tree/src/test/java/org/torproject/onionoo/server/ResourceServletTest.java#n633
 here]. And now imagine how this would look like without test method names.
 Sure, it would be shorter. But would you still understand that the input
 data for `testSearchBase64HashedFingerprintTorkaZ` is the hashed
 fingerprint of one of the example relays? Stated differently, from looking
 at the data alone, would you be able to tell whether a given test case is
 already covered by existing tests or not?

 I guess when we move to parameterized tests we should try to preserve the
 information that is currently contained in test method names. It could be
 in Java comments, though they might not be used or updated in practice. It
 could also be a String in yet another parameter (maybe first?) which might
 be printed out by JUnit in case of a failing test. Are there best
 practices for this issue, or how do others solve this?

 Another potential problem is that, AFAIU, we'll have to split each (or at
 least most) of the current test classes into one or more that share
 parameters and another one that is non-parameterized. Should be okay, but
 we should ideally have a plan for this before moving around all existing
 unit test classes.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/22733#comment:4>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online


More information about the metrics-bugs mailing list