[cknitt]

Closes #8947.

Pull requests no longer bump the .cmi, .cmt and .ast magic numbers in compiler/ext/config.ml, so parallel PRs stop conflicting on them.

• Release-time bump. yarn changelog:release bumps each counter when compiler/ml, compiler/ext or compiler/flow_parser changed since the previous release tag (git describe, excluding the tag of the version being released). These are the ml library and the libraries it depends on, so they cover every serialized type and the types they reference. A counter that already differs from the tag is left alone, so running the release step again for the same version doesn't bump twice. The bump runs after all validation and before anything is written.
• CI guard. The changelog check fails a PR without the changelog:release label that bumps a counter. Switching to another prefix is allowed.
• Own prefixes. .cmi/.cmt move from Caml1999I040/Caml1999T048 to ResCmi040/ResCmt048. OCaml's numbers count up toward the same values (OCaml 5.5 writes Caml1999I037), so OCaml tools would eventually accept our files. Both numbers stay the same length, since .cmt reading compares one header with both.
• Errors. A .cmi with a Caml1999I… header is reported as built by an older version. The message now says ReScript instead of OCaml.
• Docs. The release step in CONTRIBUTING.md, changes/README.md and compiler/AGENTS.md now describe the new process.

There is one deliberate difference from the issue text: all four numbers use the same source directories. A per-format file list (e.g. parsetree.ml only) would miss dependencies such as External_ffi_types and String_literal. Bumping too often is harmless.

For the first release after this PR, the dry run against v13.0.0-alpha.6 bumps nothing: the new prefixes and the .ast bumps made since that tag already make every number differ from it.

Existing open PRs that bump a number will conflict with this one once more. After rebasing, they should drop the bump, or the new check fails.

New tests

• scripts/magicNumbers.test.js: counter increments, bump rules (already-bumped numbers, previous release with other prefixes or without config.ml), and a temporary git repo with release tags covering unrelated changes, uncommitted changes, reruns once the version's tag exists, and the next release. Also the PR guard.
• scripts/changelog.test.js: run("release") writes the bumped config.ml.
• tests/build_tests/cmi_magic_number: new .cmi files start with ResCmi…. Legacy, older, newer and unrelated headers each get the right error.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QFCgV9acRwi6ohFaj9atR8

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.29%. Comparing base (f26ccd0) to head (090f711).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9000      +/-   ##
==========================================
+ Coverage   81.27%   81.29%   +0.01%     
==========================================
  Files         456      456              
  Lines       63834    63836       +2     
==========================================
+ Hits        51882    51893      +11     
+ Misses      11952    11943       -9     

Files with missing lines	Coverage Δ
compiler/ext/config.ml	100.00% <ø> (ø)
compiler/ml/cmi_format.ml	95.74% <100.00%> (+20.18%)	⬆️

🚀 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.

[cknitt]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-23T16:31:17.882673Z	090f711	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 93dae5cdae

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[cknitt]

@codex review

[github-actions]

Semantic Diff

4 commits · 090f711 · Oct 23, 2026, 16:25 UTC · Full page

Bump the magic numbers at release time instead of in pull requests

_{1c5a168 · Christoph Knittel + Claude · Oct 23, 2026, 15:51 UTC}

Why

Every pull request that changed a serialized type bumped the .cmi, .cmt or .ast magic number, so parallel pull requests kept conflicting on the same lines of config.ml. yarn changelog:release now bumps each counter when compiler/ml, compiler/ext or compiler/flow_parser changed since the previous release tag, and the changelog check rejects pull requests that bump one. The .cmi and .cmt numbers switch to ResCmi/ResCmt prefixes: with OCaml's Caml1999 prefix, they run into the numbers of OCaml releases (OCaml 5.5 writes Caml1999I037), and OCaml tools would take our files for theirs (#8947).

Diagnostic modified (.cmi from an earlier compiler)

diagnostic

 A.cmi
-is not a compiled interface for this version of OCaml.
-It seems to be for an older version of OCaml.
+is not a compiled interface for this version of ReScript.
+It seems to be for an older version of ReScript.

⚠️ Breaking: Tools that check for Caml1999I/Caml1999T magic numbers have to look for ResCmi/ResCmt.

Review

• Why not derive the magic numbers from a digest of the types? Tools that read .cmi files need a stable number per release that only changes when the format may have changed; a digest changes on every release and has no order.
• What about two dev builds that share a counter? Both build systems clean a package when the bsc hash changes, so only mixing files from different builds by hand reads a mismatched layout.

Type the helpers of the cmi_magic_number build test

_{36bdb7b · Christoph Knittel + Claude · Oct 23, 2026, 16:11 UTC}

yarn typecheck rejected their implicitly typed parameters.

Compare the magic numbers with the highest release below the version

_{857f88e · Christoph Knittel + Claude · Oct 23, 2026, 16:19 UTC}

Codex review: git describe took the nearest tag, so a maintenance branch merged into master made its newer tag of an older line, with other magic numbers, the previous release, and no number was bumped.

✨ Install dependencies in the changelog workflow

_{090f711 · Christoph Knittel + Claude · Oct 23, 2026, 16:24 UTC · new}

Codex review: the changelog scripts now import semver, which a clean checkout lacks.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 857f88e62a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[cknitt]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: 090f711538

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[pkg-pr-new]

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@9000

@rescript/belt

npm i https://pkg.pr.new/@rescript/belt@9000

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@9000

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@9000

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@9000

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@9000

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@9000

commit: 090f711

[github-actions]

Developer playground preview: https://rescript-lang.github.io/rescript/dev-playground/?version=pr-9000