Add native fetch() polyfill

[bkaradzic-microsoft]

Implements the WHATWG etch() API as a native polyfill under Polyfills/Fetch/, mirroring the XMLHttpRequest polyfill layout. Like XMLHttpRequest, it is built on the same platform-specific UrlLib transports, so libcurl/WinHTTP/etc. behavior is identical between the two.

Closes#98.

API

fetch(input, init) returns a Promise that resolves to a Response-like object exposing:

• ok, status, statusText, url, redirected, type, bodyUsed
• headers with get(name) / has(name) / forEach(cb) (case-insensitive)
• text(), arrayBuffer(), json(), blob() (each returns a Promise)
• clone()

Behavior / deliberate limitations

• Per the fetch spec, the promise rejects only on transport-level failures; a completed request with a non-2xx status (e.g. 404) still resolves with ok === false.
• The body is fully buffered before the promise resolves, so the body accessors may be called more than once (bodyUsed stays false) — a lenient deviation from the spec's single-use semantics.
• GET/POST only and string request bodies only, matching the underlying UrlLib transport (same constraints as XMLHttpRequest).
• blob() requires the Blob polyfill to be initialized.

Wiring

• New JSRUNTIMEHOST_POLYFILL_FETCH option (default ON). UrlLib is now fetched when eitherXMLHttpRequest or Fetch is enabled.
• Initialized in the test harness and covered by a new describe("fetch") block (12 cases: status/ok, 404-resolves, statusText, text/arrayBuffer/json/blob, case-insensitive headers, clone, init method, no-arg rejection). A small Assets/sample.json backs the deterministic json() test.

Verification

Built the Fetch target under warnings_as_errors and ran the full UnitTests suite on Win32/Chakra: 192 passing, including all 12 new fetch cases.

Context

This is step 2 of the fetch native-polyfill pivot discussed in BabylonNative#1707 (reviewers asked fetch to follow the same native pattern as File/FileReader (#169) and AbortController). Once this lands, BabylonNative#1707 becomes a small bump that initializes Polyfills::Fetch.

[Copilot]

Pull request overview

Adds a native fetch() polyfill implemented in C++ on top of the existing UrlLib transports (matching the XMLHttpRequest approach), wires it behind a new CMake option, and extends the unit test suite to validate core fetch/Response behaviors.

Changes:

• Introduce Polyfills/Fetch library implementing fetch(input, init) and a minimal Response/Headers surface.
• Wire JSRUNTIMEHOST_POLYFILL_FETCH (default ON) and ensure UrlLib is fetched when either XHR or Fetch is enabled.
• Add Mocha unit tests and a JSON asset to validate response properties and body readers (text/arrayBuffer/json/blob) plus cloning and header behavior.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
Tests/UnitTests/Shared/Shared.cpp	Initializes the new Fetch polyfill in the UnitTests runtime.
Tests/UnitTests/Scripts/tests.ts	Adds a new describe("fetch") suite covering status/ok, statusText, body readers, headers, clone, init.method, and no-arg rejection.
Tests/UnitTests/CMakeLists.txt	Links the new Fetch target into the UnitTests executable; Assets are globbed/copied so sample.json is picked up.
Tests/UnitTests/Assets/sample.json	Adds deterministic JSON payload for the response.json() test.
Polyfills/Fetch/Source/Fetch.h	Declares internal Fetch initializer.
Polyfills/Fetch/Source/Fetch.cpp	Implements fetch(), Response-like object, and headers helpers over UrlLib.
Polyfills/Fetch/Readme.md	Documents supported surface area and known limitations.
Polyfills/Fetch/Include/Babylon/Polyfills/Fetch.h	Adds public API header for initializing the polyfill.
Polyfills/Fetch/CMakeLists.txt	Defines the Fetch library target and links against JsRuntime, arcana, and UrlLib.
Polyfills/CMakeLists.txt	Adds conditional inclusion of the Fetch subdirectory.
CMakeLists.txt	Adds JSRUNTIMEHOST_POLYFILL_FETCH option and updates UrlLib dependency gating.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bkaradzic-microsoft]

Addressed both review comments in ebe6a80:

• blob() on JSC/JSI: switched the Blob-availability check from IsFunction() to IsUndefined()/IsNull(), matching the existing File polyfill workaround. Some JavaScriptCore/JSI builds report constructor functions as typeof 'object', which was causing the \�lob()\ test to fail across all JSC/JSI CI jobs.
• : added the explicit include for \std::tolower/\std::toupper.

This should turn the failing JSC/JSI/Android/Ubuntu/macOS/iOS builds green (they were all failing solely on the blob() test).

[bkaradzic-microsoft]

CI is now fully green (23/23 ✅). Pushed four follow-up fixes after the initial submission, all surfaced by platform-specific CI:

1. *\40f5ff7* — Fixed a stack-use-after-return crash (\std::system_error: Invalid argument\ on clang/libc++, JSC, JSI, Android). arcana's \ ask::then()\ captures the scheduler by reference and runs it on the UrlLib worker thread after \etch()\ has returned, so the stack-local \JsRuntimeScheduler\ dangled. Now heap-allocated and co-owned by the continuation (mirrors how XMLHttpRequest keeps its scheduler alive as a member). Caught by ASAN.
2. *\�be6a80* — \�lob()\ now detects the Blob polyfill via \IsUndefined()/IsNull()\ instead of \IsFunction()\ (JSC/JSI report constructors as typeof 'object'), plus added <cctype>. (addresses both Copilot review comments)
3. *\44bc77a* — Made the \�lob()\ \parts/\options\ Napi objects non-const (Node-API-JSI declares \Set()\ non-const → C2663).
4. *\1118799* — Linked \Fetch\ into the Android \UnitTestsJNI\ target, which has its own polyfill link list.

Ready for re-review.

[bghgary]

[Reviewed by Copilot on behalf of @bghgary]

Clean polyfill. One correctness fix and one optional cleanup inline.

[bkaradzic-microsoft]

Thanks for the review, @bghgary — all comments are addressed in ac2019f, and the follow-ups you flagged are now resolved/tracked.

Correctness fix (the blocker): the SendAsync() continuation is split into two chained .then()s — the first does the work and resolves, the second inspects the expected result and rejects on error. So a throw inside the continuation (e.g. from BuildResponse) now settles the promise as a rejection instead of being captured into a discarded result and dropped, which previously hung await fetch(...). The scheduler is co-owned by the second continuation so it outlives the in-flight request.

Cleanups: single catch (...) via the exception_ptrNapi::Error overload (removes the Napi::Error-vs-std::exception prefix asymmetry); non-allocating EqualsIgnoreCase(string_view, string_view) replacing ToLower, used by both ParseMethod and FindHeader; ResponseData::body is now a plain std::vector<std::byte>; removed the unnecessary BuildResponse forward declaration.

statusText "belongs in UrlLib" — done.BabylonJS/UrlLib#30 is merged and exposes UrlRequest::StatusText() (wire reason phrase where available, else a canonical code→text fallback table). The stacked follow-up #195 drops the private table here and reads request.StatusText() into Response.statusText (and gives XMLHttpRequest.statusText for free). Tracked + closed via #193.

Node-API-JSC napi_typeof bug (the blob()IsFunction() workaround): root cause is napi_typeof deciding function-ness solely via JSObjectIsFunction, which returns false for JSObjectMakeConstructor constructors on some JSC builds, so it reports napi_object where V8 returns napi_function. Tracked in #194.

CI is green. Re-requested your review — PTAL.

[bghgary]

[Reviewed by Copilot on behalf of @bghgary]

LGTM