[Webkit-unassigned] [Bug 53756] Refactor and/or split and/or rename the extend-selection tests to be clearer what they're doing, and able to be run manually

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=53756





--- Comment #7 from Benjamin (Ben) Kalman <kalman at chromium.org>  2011-02-03 22:47:53 PST ---
(In reply to comment #6)
> (In reply to comment #4)
> > > Making tests manually runnable is good but all these rewrites of script isn't that great.  It's really hard to see that new tests are testing the same things the old tests were testing.  In general, we must be careful when modifying tests because subtle differences may affect what tests do.  At least, we should separate refactoring and changes to make it manually runnable.
> > 
> > If you look at the changes to the expected results it's pretty clear that it's testing the same thing.
> 
> Sure expected tests are same but extend-selection-* tests relies heavily on scripts so I wouldn't modify like this all at once.

I don't think I know what you mean?  What would the value in splitting up a change like this?  Changes to the script are necessary, and whether they're made all at once or incrementally the end result is the same, that is, there would be the same amount of doubt that they're still reporting correctly.

And I think the chances of them being different are quite low given
 - they're the same now
 - for any incorrect cases, the code is simpler now.

> 
> > > I don't think separating the line boundary tests into LTR/RTL tests is a good idea because having both results on one test has helped us catching regressions in either result since extend-selection-* tests are written so that LTR/RTL results should match.
> > 
> > The reason I split those into LTR and RTL tests is because it's impossible to run them manually without doing that.  The transition from LTR to RTL actually modifies the dom mid-test.
> 
> We need to find a better way of testing both LTR and RTL cases then.  I don't think it's acceptable to separate LTR/RTL tests; we'll have regressions in no time if we did that.

We could have both the LTR and RTL doms in the same test, but when I did that there were ~50 divs on the page and it becomes more difficult to spot what actually went wrong.  It's a tradeoff I guess.

I still don't know how having them in separate tests implies regressions, though.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.