Comment 1 for bug 837496

Interesting, Stephane. I haven't seen a join fail like this before.

One issue might be the join-order that the optimiser works out for your join, likely it'll do t1 first and that might create a problem, as looking up origid makes no sense, it should be "stuffed in to" the engine so it can do its calculations. The subquery does something similar.

Do you have EXPLAIN output for this, and possibly a debug log so we can see what the optimiser is thinking?
thanks