What is supposed to happen in that case? Can the attach fail, if the attached-to DB is already locked by another TX? I have an SQLite-backed server, which shards data (*by-project*) across DBs, and when I start a transaction, I cannot know in advance all DBs involved in the transaction (start-tx and writes are in separate client messages), and I don't recall reading anything in the doc about that particular corner-case, which happens to be pretty important to me, I now realize. I could well have missed it in the doc though... specific doc pointers welcomed. I could test it I guess, but *current-behavior* is not the same as *documented* and *supported* behavior, thus this question (to Richard and Dan mostly). I'm still in JOURNAL mode, but plan to switch to WAL soon, in case it matters.