feat(js): support nested conditional exports in package.json

[akapug]

Summary

This PR adds support for nested conditional exports in package.json, fixing #1793.

Problem

GraalJS's NpmCompatibleESModuleLoader throws "Unsupported package exports" when it encounters packages with an exports field. This prevents importing many modern npm packages that use conditional exports, including:

• @discordjs/collection
• @discordjs/rest
• @discordjs/ws
• discord.js
• Many other packages following the Node.js conditional exports specification

Example failing pattern:

{
  "exports": {
    ".": {
      "import": {
        "types": "./dist/index.d.mts",
        "default": "./dist/index.mjs"
      },
      "require": {
        "types": "./dist/index.d.ts",
        "default": "./dist/index.js"
      }
    }
  }
}

Solution

Intercept npm package resolution in ElideUniversalJsModuleLoader and handle the exports field according to Node.js specification before falling back to GraalJS.

Implementation

PackageExportsResolver.kt (~220 lines)

• Implements Node.js exports resolution algorithm
• Recursively resolves nested conditional exports
• Supports conditions: import, module, default
• Handles scoped packages (@scope/package)

ElideUniversalJsModuleLoader.kt (modified)

• Added resolveWithExportsFallback() method
• Tries exports resolution for bare specifiers before GraalJS fallback
• Falls back gracefully if exports can't resolve

Supported Patterns

Pattern	Example	Status
Simple string	"exports": "./index.js"	✅
Flat conditional	"exports": { "import": "./index.mjs" }	✅
Nested conditional	"exports": { ".": { "import": { "default": "./index.mjs" } } }	✅
Subpath exports	"exports": { ".": "...", "./sub": "..." }	✅
Scoped packages	@scope/package	✅

Testing

Added comprehensive tests in JsPackageExportsTest.kt:

1. testSimpleStringExports - Basic string exports
2. testFlatConditionalExports - Flat import/require conditions
3. testNestedConditionalExports - The main fix for nested conditions
4. testScopedPackageNestedExports - Scoped packages like @discordjs/*
5. testDefaultFallback - Fallback to default condition

Design Decisions

1. Minimal changes: Only intercepts when necessary, falls back to GraalJS for everything else
2. ESM-first conditions: Prioritizes import > module > default for ESM imports
3. No breaking changes: Packages without exports work exactly as before
4. Recursive resolution: Handles arbitrarily nested conditions

Related

• Fixes Nested conditional exports in package.json not supported #1793
• Related GraalJS limitation: GraalJS throws on any exports field, not just nested ones

---

This PR was created with assistance from an AI coding assistant.

[akapug]

@darvld - Requesting your review when you have time. This enables importing npm packages that use nested conditional exports (like @discordjs/*).

Happy to make any changes based on your feedback!

[akapug]

Fixed the isDirectory() method calls - pushed 8632442. CI should pass now.

[akapug]

Fixed trailing whitespace in 0e5d19d. Formatting check should pass now.

Will investigate the test failure next.

[akapug]

Fixed syntax errors in 5cea539:

1. @discordjs/* in KDoc was being parsed as opening a /* comment - escaped with backticks
2. Removed unnecessary is String check (API returns TruffleString, not String)

[akapug]

Fixed KDoc issue in 09f26ff - backticks don't escape /* from Kotlin's comment parser, had to remove the /* entirely.

[akapug]

Kotlin code now compiles successfully ✅

The remaining CI failure is unrelated to this PR - it's a pre-existing issue with the llama-cpp-2 Rust crate:

error[E0308]: mismatched types
  --> llama-cpp-2-0.1.126/src/lib.rs:416:35
       expected `*const i8`, found `*const u8`

This is in the buildRustNativesForHost task for the local-ai package, not the JS module resolution code in this PR.

[akapug]

Update: Confirmed this llama-cpp-2 Rust error is affecting other PRs too - checked feat/crypto-random-int (run 19684791019) and it has the identical failure:

error[E0308]: mismatched types
  --> llama-cpp-2-0.1.126/src/lib.rs:416:35

Main branch runs show as "skipped" so they don't hit this. Looks like a CI-wide issue with the Rust toolchain or crate version on the test runners, not specific to any PR's changes.

[akapug]

Filed #1796 to track the CI-wide llama-cpp-2 Rust build failure.

[akapug]

Quick note on PR source

This PR was created from akapug/elide (fork) rather than directly from elide-dev/elide. This was unintentional - @akapug has write access to the main repo and intended to push branches directly.

Options:

1. Leave as-is (functionally identical, just shows fork as source)
2. Close this and recreate from a branch pushed directly to elide-dev/elide

@darvld @sgammon - any preference? Happy to recreate if cleaner repo history matters.

---

Suggested AI rules guardrail to prevent this in future sessions:

When contributing to repositories where the user has write access,
push branches directly to the upstream repo rather than creating
forks. Check `gh repo view --json viewerPermission` to verify access
before defaulting to fork workflow.

—windsurf cascade opus 4.5