microsoft / microsoft/TypeScript
Use of intersection type leads to extremely slow typecheck times
Nobody has claimed this yet.
- Dominant language
- Go
- Stars
- 111k
- Forks
- 14.3k
- Avg merge
- 1d 19h
- Merged PRs (30d)
- 117
Description
🔎 Search Terms
"intersection", "performance", "typecheck", "subtype", "time"
🕗 Version & Regression Information
- This is the behavior in every version I tried, and I reviewed the FAQ for entries about performance
⏯ Playground Link
No response
💻 Code
// Your code here
🙁 Actual behavior
I have a pretty complicated library: https://github.com/xixixao/convex-ents (it's similar to Prisma)
(I'm gonna describe the issue in-situ, because essentially the whole library is involved to get the times I'm observing)
All the code I'll mention is in this file: https://github.com/xixixao/convex-ents/blob/edgebitfaster-more-deduped/src/functions.ts
In it I have two intersection types, Ent and EntWriter. I have to use an intersection, because I want to add methods to a POJO (I do this via an intersection between a class and a Record type).
One type should be a subtype of the other (EntWriter should be a subtype of Ent).
The types have a method edge (from the class in the intersection), which returns a conditional type (PromiseEdge and PromiseEdgeWriter respectively), which results in other complicated types, but the subtyping relationship holds.
The library code typechecks fine, but the issue arises when the user of the library tries to assert that EntWriter<> is indeed a subtype of Ent:
const x: Ent<"sometable"> = someEntWriter;
Now TypeScript goes and tries to check the subtyping relationship, starting with the intersection, walking to the conditionals, and to all their branches, and to all the methods of those types, etc. It took minutes.
I can observe the slowdown by reverting PromiseEdgeWriter to match PromiseEdge exactly, and then changing one branch at a time.
After reading the performance wiki and getting a trace and going through the types, I have optimized this somewhat. I have pulled out the conditional logic into a type that is reused (see PromiseEdgeResult). This helped a lot, but the typechecking still takes 12 seconds instead of sub-second without this edge type change.
I tried pulling out the intersection to a helper type as well (EntBase) but this had no impact. It seems that the intersection is "inlined"?
🙂 Expected behavior
What I think would help me is if:
A subtype of B
X' = X<A, C>
X'' = X<B, C>
implied that X' subtype of X''
even when X<T, U> = T & U
But this doesn't seem to be what the typechecker actually assumes? (probably because intersection merged properties as per wiki)
But could I hint somehow to the compiler to not go through this whole huge reasoning chain? I know that EntWriter is subtype of Ent when their generic arguments are the same.
The intersection types themselves are ok, but it's the interplay with the two classes. The slowdown is from the edge method being checked. So alternatively I'd like to tell TypeScript: "Hey, this method does correctly override the parent class's method, you don't need to check that if the generic arguments to it are the same"
Additional information about the issue
No response
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the linked convex-ents src/functions.ts file and reproduce the EntWriter-to-Ent assignment that triggers the slowdown. Compare typechecking with the original PromiseEdgeWriter conditional branches and the optimized PromiseEdgeResult version, using compiler traces or timing to isolate the edge method and intersection checking. Done means identifying a reproducible TypeScript typechecker bottleneck and establishing a validated improvement or documented limitation.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- typescript
- Domain
- compilers, performance
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100