Does it happen much? I’m still trying to find someone who gets it a lot who can do some debugging like in
Fatal error: Detected non-empty blocksets with no associated blocks (and below that, and also earlier on).
Basically, a series of databases (fairly easy to set up) plus a huge profiling log (got disk space?) may help.
Thanks for confirming it can do this, though I don’t know if it’s better or worse. There were some fixes for CheckingErrorsForIssue1400 and FoundIssue1400Error test case, analysis, and proposal #3868 mentioned above. There might still be a gap left, but I haven’t been able to hit it when testing for it.
If you’re on Windows, the gap can be removed by using snapshot-policy for VSS, per comment earlier. Forum response to that indicates that it worked for at least one person. Maybe theoretical hole should be closed whether or not any lab testing (such as mine) can get it. Or at least current code examined more.
There would be lot less guessing if there were one or more DB bug reports to review, and some logs too.