Calls from loops run inside the loop's own machine code
A speed release for the engine's compilers. A loop that calls a small function over and over now carries that function inside its own machine code, including small wrappers that pass an array the page keeps. Code with try and catch blocks now runs in the fastest tier. Recorded real pages spend about the same time in script as in 0.1.420: one encyclopedia page is 0.6% faster, and one news page is about 0.5% slower, the expected cost of compiling its error handling.
Small functions called from loops cost almost nothing
Pages often call a tiny helper thousands of times from one loop: a wrapper that hands an array to another function, a getter, a counter. Each call used to leave the loop's machine code, set up a call and come back. The engine now copies the helper, and the function it calls, into the loop's own code, so the loop runs straight through. If the helper is replaced, or its array changes in a way the copy did not expect, the loop falls back to a normal call and gives the same answer.
try and catch run in the fastest tier
A function with a try and catch block used to stay in a slower tier. It now compiles in the optimizing tier, and a throw caught in the same function stays in machine code. Small functions start on the quick tier and move up only once they are busy, and very large ones are never given the slow optimizing compile, so a page's start-up code does not pay for it.