Comment 5 for bug 796647

Had a look at this one. I think it is still manifesting somehow.

I cant repeat it exactly as shown here; in basic.test I add and delete rows at various points so just now added some extra selects at various points to check the added data is as expected.
One condition that must hold is select count(*) from graph_base == select count(*) from graph.

But then I had an odd result in a final select * from graph after deleting some rows - from a 23 row graph and having deleted two rows then running a dijkstras query then doing a select *, the data returned from graph only returned the first 6 rows of what should have been 21... so something causes the return from `select * from graph` to not show everything after some combinations of modifications to the underlying graph.

I'll try and isolate a specific test separately from basic.test for regression testing