Tracking Substitution Type Replacement

#43427
#43425
#43439

• We create "substitution types" in conditional types - when writing T extends U ? TrueBranch : FalseBranch, we substitute T with something that's sort of equivalent to T & U in the TrueBranch.
• Doing a "dumb" substitution.
• When did we start supporting substitution types on primitives/literals?

  • Turns out it's relied upon quite a bit!

  • If you write "foo" extends keyof T ? ... : ..., you want to make a substitution type of "foo" and keyof T so that you can index into T with it in the true branch!

  • Example:

    // Ensure the following real-world example that relies on substitution still works
    type ExtractParameters<T> = "parameters" extends keyof T
    // The above allows "parameters" to index `T` since all later
    // instances are actually implicitly `"parameters" & keyof T`
        ? {​​
            [K in keyof T["parameters"]]: T["parameters"][K];
        }​​[keyof T["parameters"]]
        : {​​}​​;

• You can figure that in contravariant positions, you don't actually want to produce the substitution type.
  • The issue is that not every use of a type implies strict variance. You could end up with some invariant and covariant and contravariant uses.
• Should you create unions in contravariant positions?
  • Hard to reason about - leave it alone for now.
• PR seems right, get some more eyes on it.

Fixing Excessive Stack Depth Issues

#43249

• Lot of people hitting this issue.
• Tracks back to Add missing relationship allowing a type to be assignable to a conditional when assignable to both branches #30639
• Came up with a targeted fix that special-cases arrays, but doesn't fix everything.
• Unclear if there's a potential better way to write FlatArray. Will experiment.

Performance Depends on Declaration Order

#43437

• If you have a const that uses an interface before it's declared...it's somehow slower!
• The spread is also suspicious - spreads can create huge anonymous types.
• Potentially related to lower type IDs.
  • Subtype reduction picks the one with a lower type ID, but that could be worse to compare against.
  • Also, might just not be hitting the relationship cache.
• Investigation: start at high number, go down.
  • It didn't work. 😱
• Unclear what to do about it though.

Decorators Metadata Compat for SES

#43463