Recursion gets its speed back, and Ask reads the room

Two things this release: the engine does more work in less time, and the browser understands your reading a little better. First, speed. The part of the engine that speeds up hot code already handled tight loops, but a whole category was slipping through -> functions that call themselves. The classic case, working out a Fibonacci number the recursive way, ran no faster sped-up than plain, because every single call was quietly allocating a fresh scratch pad and throwing it away, millions of times over. Now the whole chain of calls shares one, and that computation runs about eight times faster than the plain interpreter. Building objects inside a loop -> the everyday 'turn this data into a list of records' shape -> now gets sped up too. None of it changes a single answer: the fast path is proven to agree with the plain interpreter on every program a fuzzer throws at it, and the whole engine gate still holds at 6375 of 6375. Worth saying plainly, because it is unusual: this speed-up is not a machine-code compiler, so it never needs memory that is writable and runnable at the same time -> the single biggest source of security holes in the browsers that do compile to machine code. Fast, and safe by construction. Second, memory. Ask your reading -> the on-device box that answers questions from pages you have actually read -> now pays attention to a page's title, not only its body. A page literally titled for your question used to rank no higher than one that mentioned the word once in passing. Now the on-topic page wins, the way you would expect it to.