We know that langpacks are not deterministic and are registered asynchronously. That means that it is not impossible for startup code to request strings before langpacks are registered which will result in wrong language resources being used. In L20n world, all the declarative code (XUL/HTML etc.) can retranslate on fly, but any requests for strings that are manual should not resolve until langpacks are registered. If we allow for synchronous string retrieving, we should have a test that: 1) Installs a langpack 2) Switches language to the langpack 3) Starts an instrumented browser 4) If the synchronous `getString` API is called before langpack is installed, report error.