The Perfect Follow-Up to Dictionary Performance - mapKeyedValues (SE-0510)

[Okariuss (Okan)]

Before WWDC26, I published an article called “Advanced Dictionaries: Performance and Hidden Traps.” In that article, I explained why mapValues(_:) is very fast because it reuses the same hash structure and memory slots. However, we had a big problem: mapValues does not give us the keys inside the closure. To use keys, we had to choose slow methods like reduce(into:) which cause a lot of rehashing.

At WWDC26, Apple solved this problem with Dictionary.mapKeyedValues (SE-0510).

I wrote a new follow-up article that connects my previous dictionary performance notes with this new API. In this short post, I talk about:

• A quick recap of older performance traps like uniqueKeysWithValues.
• How mapKeyedValues lets us use keys as context with less performance tax.

You can read the article here

I would love to hear your thoughts on this new API and my article

[ibex]

SE-0510 is still using the name mapValuesWithKeys, despite this:

@scanon

One commenter pointed out that because the arguments to the closure body are (Key, Value), the name mapKeyedValues may be more clear at the use site. The LSG agrees with this, and has decided to accept this alternative name.
...
Our decision is that the proposal should be accepted with those modifications. Thanks to everyone for your participation in the review.

– Steve Canon

mapKeyedValues or mapValuesWithKeys?

@scanon, @taylorswift: which one?

[Jon_Shier]

It's supposed to be mapKeyedValues, but it doesn't look like it has shipped with Xcode 27b2 yet, so I can't confirm.

[scanon (Steve Canon)]

mapKeyedValues.

The implementation PR is still open; Nate and Alex requested changes two weeks ago, which I think @taylorswift has made, so we should be pretty close to landing it.