[Win32][Edge] Fix BrowserFunction race condition using AddScriptToExecuteOnDocumentCreated

[HeikoKlare]

Summary

Fixes #20 (see also #2449 for a reproduction snippet).

In Edge/WebView2, execute() is asynchronous. When new BrowserFunction() is called, the injection script is queued via ExecuteScript(), but if a page navigation completes before WebView2 processes that queued script, the function is unavailable in the new document. This race is most easily triggered when two Browser instances are created in quick succession, as the second browser's initialization forces event-loop processing that can advance the first browser's navigation past the injection window.

Fix: Override createFunction() in Edge to additionally register the function script via AddScriptToExecuteOnDocumentCreated. This WebView2 API guarantees the script runs on every future document creation, before any page scripts — eliminating the race condition. The script ID returned by the async registration is stored so it can be cleaned up via RemoveScriptToExecuteOnDocumentCreated when the BrowserFunction is disposed.

When createFunction() is called from within a WebView2 callback (inCallback > 0), blocking on the registration would deadlock, so the persistent registration is skipped and the existing NavigationStarting re-injection remains as a fallback for that case.

Changes

• ICoreWebView2.java: Add missing RemoveScriptToExecuteOnDocumentCreated vtbl binding (index 28, between AddScriptToExecuteOnDocumentCreated at 27 and ExecuteScript at 29)
• Edge.java: Override createFunction() to persistently register via AddScriptToExecuteOnDocumentCreated; override deregisterFunction() to call RemoveScriptToExecuteOnDocumentCreated on disposal
• Test_org_eclipse_swt_browser_Browser.java: Two regression tests (isEdge-only):
  • test_BrowserFunction_availableBeforePageScripts_issue20: registers a function, navigates to a page whose inline <script> immediately calls it, asserts the function was invoked (tests the AddScriptToExecuteOnDocumentCreated guarantee directly)
  • test_BrowserFunction_availableOnLoad_concurrentInstances_issue20: creates two browsers concurrently without waiting for initialization, verifies both functions are callable when their pages complete loading (replicates the original bug report timing)

Test plan

• Run test_BrowserFunction_availableBeforePageScripts_issue20 on Windows with Edge
• Run test_BrowserFunction_availableOnLoad_concurrentInstances_issue20 on Windows with Edge
• Run existing test_BrowserFunction_* tests to verify no regressions
• Manually verify with the reproduction snippet from #2449 comment

🤖 Generated with Claude Code

[github-actions]

Test Results

  141 files   -    29    141 suites   - 29   25m 11s ⏱️ - 1m 51s
4 683 tests +    4  4 660 ✅ +    2   23 💤 +2  0 ❌ ±0 
5 519 runs   - 1 073  5 371 ✅  - 1 066  148 💤  - 7  0 ❌ ±0 

Results for commit a7f51b1. ± Comparison against base commit e476728.

♻️ This comment has been updated with latest results.

[HeikoKlare]

Test plan is done. The automated tests are run via CI anyway. I also successfully executed the reproduction snippet: #2449 (comment)

@sratz can you please review this Edge enhancement?