On 23-11-15 15:23, Richard Hipp wrote: > On 11/23/15, Johan W. Van Ooijen <jwvanooijen at kyazma.nl> wrote: >> On 23-11-15 15:01, Richard Hipp wrote: >> > Try compiling with -DSQLITE_MEMORY_BARRIER="" >> >> This will produce an empty sqlite3MemoryBarrier() function; is such a >> memory barrier not necessary under 32-bit and/or 64-bit Windows? >> > > The necessity of a memory barrier is debated. SQLite seemed to do > fine without one for 14 years. Probably you will have no problems as > long as you invoke sqlite3_initialize() and let it run to completion > from a single thread before you do anything else with SQLite in any > other thread. >
Thanks. Memory barriers are beyond my expertise, I have to rely on others about this. My software runs lots of parallel threads that each can write to the same database table. Using sqlite3_mutex_enter() and sqlite3_mutex_leave() in SQLite 3.8.5 (which has no memory barriers) this worked fine, so probably you are right. Currently I am in touch with Embarcadero to see what they can do. Memory barriers are implemented somewhere inside C++Builder, so it may be possible to enable them. Will inform you if they solve things. Did you notice my comment about the too early definition of the macro SQLITE_OS_WIN in the amalgamation? Regards, Johan Van Ooijen --
