Design Meeting Notes, 8/20/2021

[DanielRosenwasser]

Track Source and Target Stack Depths Separately

#41821

• Started around convo on Negative cases of getNarrowedType that match the exact type should be filtered out, even when generic #45205, Track source and target relationship stack depth seperately, only increase on change in value #41821 is the "better" fix.
• Put up some changes to the second pull request.
• Still waiting for DT results to come back.
• Changed the implementation to track whether or not isRelatedTo is diving into the left or right side.
• Track source and target relationship stack depth seperately, only increase on change in value #41821 (comment)
  • Lots of types have constraints and we have to "drill in" to constraints to see if the constraint is related to the target type.
  • Those really shouldn't count towards another level of comparisons when it comes to the recursion limiter - otherwise we'll often incorrectly say the types are too deep to continue comparing.
  • Similar behavior occurs for union types because unions add another layer of comparisons.
• Perf is also improved - a 1.5% improvement in Material UI.
• Feels like there is some strangeness around conditional types.

Array.prototype.at Returning undefined

#45512

• Can still index outside the range, just also works with negative indexes.
  • Could make that work on tuples, but need more special support.
  • Or, could just bake a hack in for tuples.
    • ...unless you have rests and spreads!
      • ...unless those are front-loaded! (e.g. [...some, ...stuff, 123, true, 'a'].at(-1) could return string)
  • Unclear if we would want to get into the domain of adding type notation support for it.
• We're fans of making sure this thing has undefined in the return type.
• Do we need to have the special tuple support before we ship this?
  • Possibly.
• Seems strange - in Python, experience was using regular arrays, and -1 was the argument that was passed in 99% of the time.
• 2 things we should do
  • Ship the thing that returns T | undefined in lib.d.ts.
  • Bake in some support for negative indexes when we know better.

[MartinJohns]

(e.g. [...some, ...stuff, 'c', 'b', 'a'].at(-1) could return -1)

Am I missing something here? How could this possibly return -1?

[DanielRosenwasser]

Thanks! The original code snippet used [...some, ...stuff, -3, -2, -1] but I switched the contents of the array to make it clearer.

[DanielRosenwasser]

Even that was a little bit off - the real example should use a heterogeneous array and not discuss literal types. I guess taking notes during the discussion can throw me off a bit. 😅