chore(bidi): reimplement RawKeyboard.sendText()

[hbenl]

Bidi doesn't have a protocol method corresponding to CDP's Input.insertText, so this was emulated using keyboard actions. However, this emits key* events that are not expected for Keyboard.insertText(), breaking several tests.
This implementation tries to emulate CDP's Input.insertText behavior more closely.

Fixes the following tests:

• page/page-fill.spec.ts:
  • "should not double-fill in contenteditable with beforeinput handler in Firefox"
• tests/page/page-keyboard.spec.ts:
  • "should send a character with insertText"
  • "insertText should only emit input event"

[yury-s]

Looks good except for putting Bidi-specific script into a separate extension.

This implementation tries to emulate CDP's Input.insertText behavior more closely.

Are there any plans to have such method in Bidi protocol?

[yury-s]

Since this is only used in Bidi, it should go in its own script/module. You can call browserContext.extendInjectedScript() to add extra logic.

[hbenl]

@yury-s I tried using browserContext.extendInjectedScript() but it breaks the test "should avoid side effects upon disconnect". Even without this PR, after adding

promises.push(this.extendInjectedScript(`module.exports = { default: () => class C {} };`));

to BidiBrowserContext.initialize() or to CRBrowserContext.initialize() (verifying that this is not a Firefox/Bidi specific issue), the test fails.

[yury-s]

can we avoid adding this to the main script? packages/injected/src/recorder/pollingRecorder.ts adds a rich API and doesn't leak any of it into the main InjectedScript, let's employ the same approach here.

[hbenl]

OK, I'll do that.

[yury-s]

@yury-s I tried using browserContext.extendInjectedScript() but it breaks the test "should avoid side effects upon disconnect". Even without this PR, after adding

promises.push(this.extendInjectedScript(`module.exports = { default: () => class C {} };`));

to BidiBrowserContext.initialize() or to CRBrowserContext.initialize() (verifying that this is not a Firefox/Bidi specific issue), the test fails.

Hmm, doesn't fail for me when I run the test with this change (in chromium). Are you sure there are no other changes?

[hbenl]

Hmm, doesn't fail for me when I run the test with this change (in chromium).

I found the root cause (see #41706) and it's a very subtle race condition - the timing needs to be just right to trigger it, so it's not surprising that it's hard to reproduce (it's rather surprising that it reproduces so well on my machine, although I could get the test to pass by adding a delay of 1 millisecond to it).

[github-actions]

Test results for "MCP"

2 failed
❌ [chromium] › mcp/http.spec.ts:104 › http transport browser lifecycle (isolated) @mcp-ubuntu-latest-chromium
❌ [firefox] › mcp/http.spec.ts:104 › http transport browser lifecycle (isolated) @mcp-ubuntu-latest-firefox

7733 passed, 1248 skipped

---

Merge workflow run.

[github-actions]

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/chromium/chromium.spec.ts:371 › should produce network events, routing, and annotations for Service Worker `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/popup.spec.ts:260 › should not throw when click closes popup `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/video.spec.ts:717 › screencast › should work with video+trace `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/video.spec.ts:717 › screencast › should work with video+trace `@chromium-ubuntu-22.04-node22`

49512 passed, 1161 skipped

---

Merge workflow run.