Add no-argument String.split whitespace tokenization

[ATX24]

The error

Calling String.split() without a separator currently fails arity validation instead of tokenizing whitespace:

function ReverseWords(s: string) -> string {
  let tokens = s.split()
  let reversed = tokens | reverse()
  reversed | join(" ")
}

Current diagnostic from engine/baml-compiler/src/thir/typecheck.rs:

Function baml.String.split expects 2 arguments, got 0

The explicit-separator workaround also produces incorrect tokens for mixed whitespace:

function Tokens() -> string[] {
  let s = " \thello  \nworld\r\nBAML "
  s.split(" ")
}

It only splits literal spaces, leaving tab/newline/carriage-return whitespace attached and preserving empty fields from repeated or leading/trailing spaces.

Root cause

• engine/baml-compiler/src/thir/typecheck.rs registered baml.String.split as a two-parameter native signature: receiver plus explicit separator, and method-call arity validation rejected receiver-only calls.
• engine/baml-vm/src/vm.rs enforced each native function's fixed arity before dispatch, so a receiver-only bytecode call to baml.String.split could not reach the native implementation.
• engine/baml-vm/src/native.rs::string_split always read args[1] as the delimiter and used Rust str::split.
• engine/baml-compiler/src/thir/interpret.rs::evaluate_method_call also required exactly one delimiter argument.

The fix

• Allow s.split() through THIR typechecking while preserving s.split(separator) typechecking.
• Allow the VM to dispatch baml.String.split with one VM argument only for the receiver-only no-arg form.
• Update string_split to use Rust split_whitespace() when called with only the receiver, which collapses contiguous whitespace and omits empty tokens; explicit separators continue to use str::split.
• Update the direct THIR interpreter path to support both zero and one split argument.
• Add Rust tests for no-argument whitespace splitting and typechecking.

Verification

Passing commands:

$ mise exec -- cargo test -p baml-vm string_split --test strings -- --nocapture && mise exec -- cargo test -p baml-compiler typecheck_string_split_without_separator --lib -- --nocapture
...
test result: ok

$ mise exec -- cargo test --lib
...
test result: ok

$ mise exec -- cargo test --features skip-integ-tests
...
test result: ok

The same reproduction now succeeds via the VM test added in engine/baml-vm/tests/strings.rs:

function main() -> string[] {
  let s = " \thello  \nworld\r\nBAML "
  s.split()
}

Expected and now-passing output:

["hello", "world", "BAML"]

Full language integration runner status:

$ mise exec -- ./run-tests.sh

This reached the TypeScript integration phase and then blocked on an interactive Infisical login prompt:

No valid login session found, triggering login flow
? Select your hosting option:

Running the TypeScript integration tests directly without Infisical also failed due missing/invalid provider credentials, not this code change:

$ pnpm test -- --silent false --testTimeout 60000
...
LLM client 'GPT35' requires environment variable 'OPENAI_API_KEY' to be set but it is not
LLM client 'Sonnet' requires environment variable 'ANTHROPIC_API_KEY' to be set but it is not
LLM client 'Gemini' requires environment variable 'GOOGLE_API_KEY' to be set but it is not
Request failed with status code: 401 Unauthorized ... invalid_api_key
Test Suites: 34 failed, 8 passed, 42 total
Tests: 155 failed, 81 passed, 236 total

Issue Reference

• This PR fixes/closes #[issue number]

Changes

Implemented no-argument String.split() for whitespace tokenization while preserving explicit separator behavior.

Testing

• Unit tests added/updated
• Manual testing performed through focused VM/typechecker tests
• Tested in Cursor Cloud Linux environment

Screenshots

Not applicable.

PR Checklist

• I have read and followed the contributing guidelines
• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings

Additional Notes

The full integration suite requires authenticated provider credentials or a valid Infisical session in this environment.

  

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
beps	Ready	Preview, Comment	Jun 4, 2026 1:59am
promptfiddle2	Ready	Preview, Comment	Jun 4, 2026 1:59am

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
promptfiddle	Skipped		Jun 4, 2026 1:59am

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 7993b926-e6d4-4b5e-9b5c-970aabd7dbc4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch cursor/no-arg-string-split-8a66

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

⏭️ Performance benchmarks were skipped

Perf benchmarks (CodSpeed) are opt-in on pull requests — they no longer run on every push. They always run automatically after merge to canary/main.

To run them on this PR, do any of the following, then push a commit (or re-run CI):

• Add RUN_CODSPEED=1 to the PR description, or
• Include run-perf or /perf in the PR title or any commit message.