Ambiguous lexical analyzis due to escape character

There's yet another regression of XRENDERING-88 (http://ci.xwiki.org/job/xwiki-enterprise-test-wysiwyg/org.xwiki.enterprise$xwiki-enterprise-test-wysiwyg/1399/testReport/org.xwiki.test.wysiwyg/MacroTest/testEditMacroWithSpecialCharactersInParameterValues/). This is, however, a more general problem of ambiguity in how the escape charater, '~', is implemented:

This is a typical example:

      <#PARAMS:   "(%" ( "~" ~[] | ~["%"] | ["%"] ~[")"] )* "%)" >

But there is a problem with ambiguity here. Take this example:

(% %~~%) %)

The PARAMS token here will be suprisingly be the whole string, because the substring "%~" matches the subpatter ["%"] ~[")"], and taking that path leads to a longer token.