David Bremner writes on February 27, 2016 13:50: > Gaute Hope <eg@gaute.vetsj.com> writes: > >> >> While loading the threads, if I make enough modifications to the >> database while the query is still loading (specifically 1: removing the >> unread tag from a thread, and 2: adding the unread tag to the same thread) >> apparently the query is invalidated in some way, and I get a hard crash >> in `notmuch_tags_get ()`. >> >> The thread in question has alreaday been loaded and is displayed. >> >> Do you deal with this issue in some way? >> > > I think we "deal with" it by not leaving the database open very > long. The query is dumped as s-expr by a seperate process. Or perhaps we > just aren't doing the same level of concurrent operations, I don't know. Neither do I, I load the threads in the background. But for a long query that takes a while. The database is only kept open for the duration it takes to load the threads in the query. How long is the db open? Would you notice if it crashed? I'm guessing the abort is triggered by an Xapian::ModifiedDatabase somewhere. I am only able to consistently reproduce the error if I change the tags of a thread not yet loaded, a (somewhat messy) test case is available here: https://github.com/gauteh/astroid/blob/ti-stateless-query/test/test_notmuch_standalone.cc If nothing else, it demonstrates the difficulty of handling these exceptions in functions that don't yet catch them, since they could happen along many of the internal steps. Regards, Gaute