Probably introduced by latest changes to custom-compile-unit cache or rendertemplate.
Added as test:
in https://svn.strategoxt.org/repos/WebDSL/webdsls/trunk/test/succeed-web/local-template-redef.app:

assert(d.findElements(SelectBy.className("b")).length > 0, "span class of local redefined template b should be 'b'");

Now the html contains:
<span id="root_b1$l$Int" class="scopediv root_b1$l$Int root_b1$l$">correct<input name="9e4b753025231367b8292684ff51f0eb" value="0" class="inputInt "></span>

Submitted by Elmer van Chastelet on 11 November 2013 at 14:02

Or.. is this actually expected behaviour?
I find this code very confusing because of the stratego and java variables with the same name:
http://codefinder.org/viewFile/x_uniquename/https%3A%5Es%5Essvn.strategoxt.org%5Esrepos%5EsWebDSL%5Eswebdsls%5Estrunk%5Essrc%5Esorg%5Eswebdsl%5Esdsl%5Esto-java-servlet%5Esjava-template.str/WebDSL#179

Strange, the issue also persists when reverted back to r5815.
Given that the deployment of researchr r65 was done 25 October with most recent build at that time, I don’t understand yet why r5815 shows the same issue (the issue causes bad layout of researchr due to pointers to the span id’s in its css).
Will try
* make fullclean @ webdsl compiler
* compare local webdsl source with repo version

After make fullclean, r5815 works fine for researchr.

r5816 breaks it, question is: What is the expected value of the class attribute for a local redefined template with span modifier?

Fixed in r5848