Hello, I have a couple of questions about auto-vacuuming.

First, with auto_vacuum=FULL, the docs say that, after every transaction, SQLite will try and shrink the database by moving free pages to the end and then truncating. That sounds like it will take a write lock (and, indeed, I see in the code that doing a manual incremental_vacuum is a write operation), even after read-only transactions, which is less than ideal for concurrency.

If my understanding of this is correct, I think it should be noted in the docs for auto_vacuum and incremental_vacuum, so that you don't have to be making deductions to avoid performance traps.

Second, how does auto-vacuuming work with WAL mode? It can't just move pages around and truncate the database immediately, because there might be readers concurrently looking at the database. So the auto-vacuum needs to go through the WAL, which seems sensible for moving the free pages around, but it's not totally immediately clear when the truncation can happen - it must be after a checkpoint completes, but how does SQLite know to do it then, which is at a remove from the auto-vacuum (or incremental vacuum) operation? Does checkpointing entail truncating free pages at the end of the database if auto-vacuuming is enabled (even in incremental mode)? If so, I think that's also worth noting in the docs, though as an explanation for the curious rather than a notice for the unwary.