On Fri, Feb 7, 2014 at 2:34 PM, Nicholas Nethercote <[email protected]> wrote: > A more general question: what exactly is atomization for? Is it purely > for cases where we want to avoid duplicating strings (with the option > of interning)? The fact that lots of data structures (e.g. ParseNode) > have JSAtom* fields suggests that it's not this simple. This is > something I've never entirely got my head around.
Atoms also make string comparisons fast: two atoms have the same chars iff their pointers are equal. So if you have an obj[foo] expression, we can atomize foo, convert it to a jsid, then walk over all shapes and do a single jsid comparison to find the right one. I think the frontend also uses this for various HashMaps that have JSAtom* as key, see Parser::noteNameUse for instance. String equality is also easy for the JITs to inline when you have two atoms, see MacroAssembler::compareStrings. If we stop atomizing string literals we will probably take a slow path there, but I don't know how much that will affect our main benchmarks... HTH, Jan _______________________________________________ dev-tech-js-engine-internals mailing list [email protected] https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
