Just curious, why not simply use the same technique in both cases? The disambiguator parameter should work find for non-constructor
methods,
right?
Adding the extra parameter means fiddling with the stack to push the extra null argument. Since non-constructor method invocation occurs more often, just changing the method name means I'm changing the runtime behavior less. If you think having two mechanisms is too confusing, I can make all method invocations use the parameter-based approach. While doing more extensive testing, I found some variations in how different compilers treat invocations of private constructors from inner or enclosing classes. This shows up in the emma tests, since we use the externally-compiled classes. I've updated the disambiguator calculation to have a fallback when this happens. http://gwt-code-reviews.appspot.com/473801/show -- http://groups.google.com/group/Google-Web-Toolkit-Contributors
