[cknitt]

Part of #9256 (field access on a large object type).

Each o["f"] instantiated the whole type of o, and the .cmt file stored every copy (writing it was the largest cost), so reading each field of an n-field object once was quadratic.

When the identifier's type is generic and has no type variables, an instance would be an identical copy. o then keeps the type itself, and only the field's type is instantiated, as before. filter_method doesn't change a type that has the field, and the shared type is never unified: the identifier isn't unified with the expected type, which is a fresh variable there. Types with variables are still copied. The check walks the type once and remembers the last type asked about, as reads of one object usually follow each other.

Measured with bsc on linux-arm64, allocated words and wall time:

Fields	Before	After
400	21.1M, 189 ms	3.4M, 34 ms
800	77.4M, 867 ms	6.5M, 62 ms
1600	297M, 3.95 s	12.9M, 136 ms

What remains is the linear scan of the field list per read.

New test: undefined_method_closed_object reads an existing and a missing field of a toplevel object, so the error goes through the new path.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Kbf3gK1dcjbe97yY1wRMgs

[github-actions]

Changelog

💅 Polish

• Speed up reading many fields of a large object, and shrink the .cmt files of such modules.

Semantic Diff

2 commits · Full page

Read fields of a variable-free object without copying its type

_{426355d · cknitt + Claude · Nov 07, 2026, 10:06 UTC}

Why

Each o["f"] instantiated the whole type of o, and the .cmt file stored every copy, so reading each field of an object with n fields once cost O(n²): 800 fields allocated 77M words and took 0.87 s. They now allocate 6.5M words and take 0.06 s.

When the identifier's type is generic and has no type variables, a copy would be identical, so the identifier keeps the type itself and only the field's type is instantiated. Looking up a field that the type has doesn't change it. Types with variables are still copied.

✨ Type the object of a field read through its own identifier helper

_{41ad3b0 · cknitt + Claude · Nov 07, 2026, 19:05 UTC · new}

The benchmarks showed every expression allocating a little more (up to 0.17%), because the new optional argument of type_expect was captured by the closure it builds for each expression. Identifiers are now typed by type_ident, which the field read calls directly for a bare identifier, so type_expect is as before.

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.74%. Comparing base (bf35764) to head (41ad3b0).

Files with missing lines	Patch %	Lines
compiler/ml/typecore.ml	93.33%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #9259   +/-   ##
=======================================
  Coverage   81.74%   81.74%           
=======================================
  Files         457      457           
  Lines       65713    65729   +16     
=======================================
+ Hits        53715    53731   +16     
  Misses      11998    11998           

Files with missing lines	Coverage Δ
compiler/ml/typecore.ml	89.18% <93.33%> (+0.06%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pkg-pr-new]

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@9259

@rescript/belt

npm i https://pkg.pr.new/@rescript/belt@9259

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@9259

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@9259

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@9259

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@9259

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@9259

commit: 41ad3b0

[github-actions]

Developer playground preview: https://rescript-lang.github.io/rescript/dev-playground/?version=pr-9259

[github-actions]

🎉 Compiler benchmarks

Allocated words

In total: 456,989,425 → 439,257,959 (-3.88%).

🎉 Improvement: 1 benchmark allocates more than 1% less.

Benchmark	Base	Head	Change
stress/ObjectFields.res	21,040,664	3,296,387	-84.33%

Instructions

In total: 10,226,962,471 → 9,405,550,072 (-8.03%).

🎉 Improvement: 1 benchmark executes more than 1% fewer instructions.

Benchmark	Base	Head	Change
stress/ObjectFields.res	957,097,762	135,501,721	-85.84%

Pull request 41ad3b0 compared with base bf35764 in this run.