Hi, On 11 September 2016 at 22:59, Christoph Cullmann <cullm...@absint.com> wrote: > I think the main issue is: There is no need to reproduce them, it is clear, > that baloo will crash > on any problem with lmdb, as "no" errors are handled (or lets say 99% of the > errors are not handled). > > The code is full of Q_ASSERT, but no handling of the return codes beside that.
Oh, it gets worse. Q_ASSERTs compile to nothing in release mode so you don't even have meaningful crashes on those asserts, things just mysteriously fail elsewhere. I spent a whole bunch of time coding in actual checks after those asserts because Dolphin would crash when selecting multiple files with Baloo disabled, and other such magical bugs. > Actually, I think, in most cases the code shall just catch the error cases > and do nothing (or purge data > until all is fine again) but not like now: crash or assert. > > That is not acceptable for something that runs per default ;=) > > Just looking at the code, you see things like: > > 1) Baloo::Database: needs a mutex, as many threads might call ::open (e.g. in > krunner) => corruption, dead > > 2) inproper cleanup: I doubt one can mdb_txn_abort(txn); if already > mdb_txn_begin(...) failed > > 3) needs everywhere return code checks, e.g. in PostingList > PostingDB::get(const QByteArray& term), we have > tons of bugs about random crashs afterwards: > https://bugs.kde.org/show_bug.cgi?id=367480 > > 4) 32-bit systems supported at all? ATM, after 1GB of indexing, that was it, > no more baloo or any other application > calling any of the accessors of e.g. Query (if you have bad luck). If you have the time to fix those things you've listed above, awesome - I'd love to look at the RRs. Lessens my workload. Otherwise I'll look at this on the weekends when I have no sysadmin things to do. > >> >>> Beside that, could we get some baloo-b...@kde.org list as default assignee >>> for all baloo bugs instead of "one" person? >> >> If we're going to do it this way, why not just make it the default >> frameworks mailing list, or the frameworks bugs list if there is one? > I am not sure, if that will make people happy ;=) Let's get other people involved, too ;-) Thanks, Boudhayan
