Community
Participate
Working Groups
We should be able to handle empty files when we are restoring the workspace. For instance, if we crash in the middle of writing out the workspace files, (or markers or syncinfo or...) then the files will have length 0 when we try to read them in during the #restore. NOTES: JM (6/26/2001 3:54:28 PM) Are you saying we don't? What are the steps to recreate this? What is the failure mode? More info please. GWG (6/26/2001 5:35:51 PM) Actually DJ logged this while visiting my office last week, so he can give more details. From what I remember we crashed somehow (possibly by my own doing, not the platform's), and when we restarted there was a problem. DJ, do you remember more details? DJH (6/26/01 7:11:46 PM) I believe that the scenerio was that a full_save on exit was interrupted (process killed?) and never completed. This left an empty file where we were writing since the stream was never closed. When the workspace is restored, the restore code tries to read an int to get the version number for the reader and it fails with an EOFException which is wrapped in a ResourceException. See SaveManager.restoreTree for details. This is also applicable to restoring markers (MarkerManager.restoreFromSave) and syncinfo (Synchronizer.restoreFromSave).
PRODUCT VERSION: R0.9
Fixed. Now tolerating empty syncinfo and marker files on restore