[jsr-314-open] Handling of "" vs null action outcomes

[Andy Schwartz]

Gang -

One of our unit tests here at Oracle tripped up over a small change in 
behavior between JSF 1.2 and JSF 2. The unit test contains an action 
method that returns an empty string, eg:

<h:commandButton action="#{foo.doSomething}"/>

public String doSomthing()
{
// Do something here

return "";
}

When running against JSF 1.2, the empty string outcome would result in 
the test case remaining on the same page, as there is no navigation case 
that matches this outcome. In JSF2, implicit navigation kicks in. In 
particular, the following statement from section 7.4.2 is applied:

> If viewIdToTest does not begin with “/”, take the current viewId and 
> look for the last “/”. If not found, prepend a “/” and continue 
> Otherwise remove all characters in viewId after, but not including, 
> “/”, then append viewIdToTest and let the result be viewIdToTest.


In our case, we have the following values:

- Current viewId: "region/refresh".
- viewIdToTest: ""

The resulting view id after applying the above logic is "region/". This 
ends up being treated as an implicit view id. In our particular test 
case, there is no view associated with this view id. Instead of 
remaining on the same page, we now attempt to navigate to a view that 
does not exist.

Our test case made the assumption that returning the empty string from 
the outcome would result in the same behavior as returning null. This 
was true in JSF 1.2, but no longer appears to be true in JSF2. We have 
tweaked this unit test to return null so that our tests are now passing, 
but I wanted to follow up to find out whether this small difference in 
behavior between "" and null outcomes is intentional. It seems to me 
that these should behave the same. It would be easy enough for us to add 
a sentence to the spec that states that empty string outcomes should not 
be treated as implicit view ids. Does anyone here know of a reason why 
we shouldn't spec this?

FWIW, my feeling is that, while returning null is a reasonable 
workaround, there is no way that our unit test is not the only code out 
there that is returning empty action outcomes and expecting to stay on 
the same page. I would like to see this behavior preserved not so that 
we can revert our unit test to its previous behavior (I am fine with our 
fix), but because the current behavior may break other users as they 
upgrade to JSF2.

Thoughts?

Andy