Text-reading loops and element-tree builders run faster
A speed release for the engine's compilers and layout. Loops that read text one character at a time, the way tokenizers and parsers do, now work on whole numbers. try and catch blocks recover into machine code sooner. The optimizing compiler now builds the element trees that page frameworks make with small helper functions, and layout reads margins, padding and border widths without parsing them each time. Recorded real pages spend about the same time in script, style and layout as in 0.1.421: two news pages are 1-3% faster, and one technology news page is about 0.5% slower.
Loops that read text one character at a time
A tokenizer reads a string one character code at a time and checks each code against a list of cases. Those codes are whole numbers, but compiled code held them as general numbers, so every case was a slower comparison. Now the code, its position in the text and each case of the switch are whole numbers. If a read goes past the end of the text, the loop leaves its machine code and is compiled once more in the general form, so the answer is the same.
Element trees built in the optimizing compiler
Page frameworks build their element trees with tiny helper functions that return an object: a tag, its properties and its children. The optimizing compiler now builds these directly, including empty lists and objects, a choice between two words such as a class name, and helpers declared inside another function, which used to fail the compiled code's check on every new call.