Known bugs

• "Save rich" is always assumed to succeed, since the system asks the browser to save the document and there's no possibility of feedback. This means that if the document has been changed, the system will believe that the changes have now been saved, whether or not the user completes the save action in the browser. Future undo/redo steps will not prompt the user to confirm that unsaved changes can be undone.
• Bug 15859: backtrace is hidden in some bug reports.
• Bug 16098: application menubar is not foregrounded when popup menus are foregrounded.
• Bug 16227: rerunning the experiment engine, except to regenerate scores, seems to be a good idea which is poorly implemented.
• Bug 18422: Neither the Python nor the Javascript implementations for document manipulation do the right thing for Unicode characters outside plane 0. If your document contains such characters, expect your offsets to be wrong; the system is basically unusable for such documents, and unfortunately does not yet report that.
• Bug 21049: XML inline rendering and reading should skip the untaggable tags, which typically cause crossing dependency problems. This fix should probably involve moving these tags into a separate administrative section which doens't count as core annotation.
• Bug 23906: The Java service API only permits JSON files to be imported into workspaces.
• Bug 24027: Under some unusual circumstances involving the proxy_for_steps attribute in task XML files, the MAT engine will not be able to infer that certain steps have occurred. This problem arises when proxy_for_steps has been used to group together multiple overlapping but not identical sets of steps.
• Bug 27456: The model inventory for the Web server is collected at startup time. If you build a model after the Web server starts up, it won't see the model.