On Fri, 15 May 2015 18:34:33 +0000 Eric Hill <Eric.Hill at jmp.com> wrote:
> Thanks for your reply. > > I went as high as using (number of rows)^4, resulting in a cost of > ~440 trillion for the unindexed case, along with setting the cost to > 1 for the indexed case, and it still won't use my index. > > I'd like to step out of my xBestFilter implementation into SQLite > code to see if I can tell what is going on there, but when I step > out, the call stack knows where I'm supposed to be, but the debugger > does not find the right line in sqlite3.c. I have built sqlite3.c > simply by adding the amalgamation to my Microsoft Visual Studio 2013 C > ++ project. Is there something I can do to make the debugger work? > > I will postpone index creation until the call to xFilter, I reckon, > once I work out these other issues. Thanks for the tip! Perhaps you can disable the unindexed path, forcing to use always the index. This way your virtual table always use your index. What I don't know is: Sqlite3 can use only one index per query, I don't know if indexes from virtual tables count for this limit. I suppouse not because it should be an internal virtual table decission implementation show the selected data the faster way. If it counts for index limit, perhaps Sqlite3 found a faster index when joining your virtual table data with other table data. Perhaps I miss read something in documentation. Dr. Hipps, can you explain if virtual table index count for this limit? > Eric --- --- Eduardo Morras <emorrasg at yahoo.es>
