> On May 10, 2017, at 2:20 PM, Simon Slavin <slav...@bigfraud.org> wrote:
> I’m concerned that this sort of corruption is happening repeatedly. Whether > or not there’s anything in howtocorrupt.html that rings any bells, you can > rely on us to help figure out what’s wrong. > Thanks. That's kind. It could well be memory corruption. We've had a hard time getting Valgrind up on this platform to chase. When we turn all the debug check stuff on in sqlite it doesn't seem to happen. (In fact, it doesn't happen often in any case.) This use case on this particular DB is dead simple too -- one writer and one reader in different posix threads in the same process, fully serialized. The unusual parts of our setup are: 1. The DB in questions is on a volatile filesystem and ATTACHED to another, smaller one residing on a flash (JFFS) filesystem, and 2. We use the shady SQLITE_SHM_DIRECTORY compile switch to move the .shm file off JFFS and into RAM so we can WAL, and 3. We're linking against an older version of the MUSL c library which has had bugs in the past on our platform and may have more. (We are moving to a new platform soon so will get to upgrade and isolate this variable.) #2 is the one I always worry about, but generally it seems to work well. -- Ward _______________________________________________ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users