Hacker News Reader

Comment by pcwalton | Hacker News Reader

Objective-C uses a hash table cache for all method calls. It's a bit different in that Objective-C deals with collisions by just not handling them: since it's only a cache, collisions will result in going through the slow name lookup path.

Mike Ash has a nice series that goes through the disassembly of objc_msgSend: https://www.mikeash.com/pyblog/friday-qa-2017-06-30-dissecti...

View on HN

But aren't hash tables orders of magnitude slower than dynamic dispatch with jump tables? I'm just interested in PL implementation as a hobby, so correct me if I'm wrong, but this seems horribly inefficient to me.

You can't use jump tables in Objective-C because methods can be added to objects at runtime.

That doesn't preclude using jump tables. You can also add methods to objects at runtime in JS and modern runtimes use jump tables.

I expect you can't use jump tables in objc because you can't really "patch in" jump tables at runtime as it's AOT-compiled rather than JIT-ed.

Right, but that's no longer "just" jump tables, you need additional JIT logic for those tables to be correct.

Plus, in a JIT you can go even further and just do direct calls, no need for the indirection of tables.

well, does macOS feel like a fast operating system to you ?

Yes!

you certainly haven't used Linux on a mac then. On the exact same machine it is incredibly more snappy.

Mike Ash’s posts are gold. I wish there were more in depth technical posts like those on HN or Lobsters.

lobsters?

https://lobste.rs, discussion-based aggregator site similar to HN.

more similar to HN than I expected. Thanks, I'll check it out!

> Objective-C uses a hash table cache for all method calls.

Nit: method calls to objects that are not nil, since the latter causes objc_msgSend to bail out before the hash table is consulted.

Reply on news.ycombinator.com