Design Meeting Notes, 5/24/2019

[DanielRosenwasser]

Correct Constraint Checking

#31006

• Could park behind a flag, maybe within --strict mode
  • But silly not to give this to everyone?
• Breaks Ember's .d.ts files (@chriskrycho)
• Does this help everyone, or is this something you notice as a power-user?
• Conclusion: update PR, update user test suite, re-evaluate and update Ember if necessary

OOM Woes

#30050

• We've been receiving a lot of OOM woes.
• All seem to have the same root cause.
• Related to normalization of large intersection types.

  • (A | B | C | D) & (E | F | G | D)

    interface A { A: "A" }
    interface B { B: "B" }
    interface C { C: "C" }
    interface D { D: "D" }
    
    interface E { E: "E" }
    interface F { F: "F" }
    interface G { G: "G" }
    interface H { H: "H" }
    
    type BLAH = (A | B | C | D) & (E | F | G | H)
    
    // Equivalent to...
    type BLAH = (A & E) | (A & F) | (A & G) | (A & H) | (B & E) | (B & F) | (B & G) | (B & H) | (C & E) | (C & F) | (C & G) | (C & H) | (D & E) | (D & F) | (D & G) | (D & H)

    • This grows pretty fast.
    • Comes up a lot when you use something like keyof React.ReactHTML
      • We run out of memory at 2**19, and some operations on React have calculated 2 ** 111.
    • We now have a limiter during the normalization pass to compute the upper-bound of the number of unions we might generate, so that we gracefully fail instead of generating out-of-memory errors.
• This will generate some trouble for React users.
• Also looks like there are a lot of bogus intersections we choke up on.
  • "a" & { __tag: "adsad" }
  • Maybe there's a subset of intersections with object types.
  • We need some way of doing nominal tagging
  • Can we just determine what to do based on number of members?
    • Nope, expensive to determine this.
• Resolution: Need to put this into the breaking changes list.

[chriskrycho]

Thanks for the ping! I am happy to carve out some time next week to look at Ember's types with you folks, and will also dig into the PR in question to try to understand why it breaks our types.

[zpdDG4gta8XKpMCd]

We need some way of doing nominal tagging