David Blass

1. the itch

   I was digging into tRPC type performance, and those two calls were in AnswerOverflow, a public tRPC app- so, for once, the trace is still up. Zod schemas derived in chains like theirs are very common in the wild, especially in the repos I investigate.

2. the hunt

   I minimized it to a chain of merges, the kind Zod's .merge().pick().omit() and tRPC's merge<merge<…>> produce. Each link roughly tripled the compiler's work until, 13 links in, TypeScript declared the chain excessively deep and possibly infiniteerror TS2589: Type instantiation is excessively deep and possibly infinite.What TypeScript 5.8.2 says about 13 chained .merge() calls: the bench below, reproduced Oct 13, 2026.tsc 5.8.2 · reproducedTS 5.9.3: 0 errors. I counted it with Attest ↓

   show the bench

   bench · chained11 calls

   type merge<base, props> =
   	Omit<base, keyof props & keyof base>
   	& props
   
   type Type<t> = {
   	merge: <r>(r: r) => Type<merge<t, r>>
   }
   
   declare const 
   const a: Type<{
       a: 1;
   }>a: Type<{ a: 1a: 1 }>
   
   const 
   const c1: Type<merge<{
       a: 1;
   }, {
       b: number;
   }>>c1 = 
   const a: Type<{
       a: 1;
   }>a.merge({ b: 2 })
   const 
   const c2: Type<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>>c2 = 
   const c1: Type<merge<{
       a: 1;
   }, {
       b: number;
   }>>c1.merge({ c: 3 })
   const 
   const c3: Type<merge<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>, {
       d: number;
   }>>c3 = 
   const c2: Type<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>>c2.merge({ d: 4 })
   const 
   const c4: Type<merge<merge<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>, {
       d: number;
   }>, {
       e: number;
   }>>c4 = 
   const c3: Type<merge<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>, {
       d: number;
   }>>c3.merge({ e: 5 })
   const c5 = 
   const c4: Type<merge<merge<merge<merge<{
       a: 1;
   }, {
       b: number;
   }>, {
       c: number;
   }>, {
       d: number;
   }>, {
       e: number;
   }>>c4.merge({ f: 6 })
   const c6 = c5.merge({ g: 7 })
   const c7 = c6.merge({ h: 8 })
   const c8 = c7.merge({ i: 9 })
   const c9 = c8.merge({ j: 10 })
   const c10 = c9.merge({ k: 11 })
   const c11 = c10.merge({ l: 12 })
   const c12 = c11.merge({ m: 13 })
   const c13 = c12.merge({ n: 14 })
   const c14 = c13.merge({ o: 15 })
   const c15 = c14.merge({ p: 16 })

   TS 5.8.2⛳ Result: 1,596,004 instantiations

   TS 5.9.3⛳ Result:       976 instantiations

   TS 5.8.2 TS 5.9.3

   linearlog

   type instantiations

   0500K1M1.5M2M101001K10K100K1M10M

   ×2.995 per link

   5.9 still grows quadratically, just not 3× per link

   1,596,004TS 5.8.2976TS 5.9.3

   15101511calls

   − .merge()+ .merge()← keep going↺

   1,635×fewer instantiations at 11 calls

   Line chart. As chained .merge() calls increase from 1 to 15, TypeScript 5.8.2's type instantiations roughly triple with each call, from 98 to 4,784,888 at 12 calls, then fail with error TS2589 from 13 calls on. TypeScript 5.9.3 stays under 1,700 throughout: 976 at 11 calls and 1,632 at 15.

   Instantiation counts from @ark/attest 0.57.0: deterministic. Bench from microsoft/TypeScript#61505; re-run on 5.8.2 and 5.9.3, Oct 13, 2026.

   Data

   Type instantiations by number of .merge() calls (attest 0.57.0). "TS2589" means the checker bailed; the count it reported is in parentheses and isn't real work.

   calls	chained 5.8.2	chained 5.9.3	functional 5.8.2	functional 5.9.3
   1	98	36	30	14
   2	250	85	77	33
   3	524	144	163	55
   4	1,136	213	360	80
   5	2,734	292	884	108
   6	7,262	381	2,383	139
   7	20,552	480	6,801	173
   8	60,100	589	19,970	210
   9	178,394	708	59,386	250
   10	532,898	837	177,537	293
   11	1,596,004	976	531,887	339
   12	4,784,888	1,125	1,594,828	388
   13	TS2589 (9,785,026)	1,284	4,783,536	440
   14	TS2589 (14,785,170)	1,453	14,349,539	495
   15	TS2589 (14,785,298)	1,632	TS2589 (24,349,567)	553

3. the fix

   I worked on it with Andarist (Mateusz Burzyński), who wrote the compiler patch: cache instantiations on type mappers. His PR opens: "@ssalbdivad found out that a chain of somewhat trivial operations can easily end up with dreaded 'Type instantiation is excessively deep and possibly infinite'." I tested it on a bunch of tRPC monoreposconst trpcMonorepo: { before: "13.04s"; after: "6.00s"; runs: 1 }One of the tRPC monorepos I tested, TS 5.8 vs. the #61505 build.A single run.@see microsoft/TypeScript#61505 (comment)public recordsourced: it removed the depth errors, cut check time in half or better, or both.

   // "How has this man not been hired by the TypeScript team yet?" — me, about Andarist, 2024

4. the review

   Anders Hejlsberg's review asked to cache only the expensive instantiations. My reply: "whether or not individual instantiations are expensive, if there are O(3^N) of them, they will eventually dominate the check time of the entire type."

5. shipped

   It merged in June 2025 and shipped in TypeScript 5.9 as "Cache Instantiations on Mappers"; the release notes name Zod and tRPC. Jake Bailey ported it to the Go compiler.

6. what repos saw

   Type-checking AnswerOverflow's dashboard dropped from about 6.7s to 3.3sconst answerOverflow: { before: "6.66s"; after: "3.29s"; runs: 5 }AnswerOverflow's dashboard, type-checked on TS 5.8 vs. the #61505 build. Mean of 5 runs each; the ranges don't overlap.Jake Bailey (TS team) measured the same repo independently: 7.67s → 3.79s.@see Andarist's runs (comment on #61535)@see typescript-go#1358 (Jake Bailey's run)public recordsourced (5 runs each, no overlap), and Jake Bailey on the TypeScript team independently reproduced the halving on the same repo. At Klarna, an engineer reported their check time dropping from six and a half minutes to about twoconst klarna: { before: "395.48s"; after: "132.09s" }Klarna's app.Third-party and self-reported, in one comment.@see microsoft/TypeScript#61505 (comment)@see microsoft/TypeScript#61505 (Klarna, follow-up)public recordsourced: "This was a game-changer PR for us."

Compiler patch by Mateusz Burzyński (@Andarist) · microsoft/TypeScript#61505 · TS 5.9 notes · typescript-go#1358