315312 2026-05-21 12:29:17 -0700 Web Inspector: didCommitProvisionalPage should disable Inspector instrumentation for the old process to keep m_instrumentedProcessPageCounts accurate 2026-05-22 20:56:10 -0700 1 1 1 Unclassified WebKit Web Inspector WebKit Nightly Build Unspecified Unspecified NEW https://github.com/WebKit/WebKit/pull/65376 InRadar P2 Normal --- 314902 1 bburg bburg inspector-bugzilla-changes webkit-bug-importer oldest_to_newest 2213047 0 bburg 2026-05-21 12:29:17 -0700 Follow-up from PR #65376 (webkit.org/b/314902) review by Sihui Liu (szewai): > didCommitProvisionalPage doesn't call disableInstrumentationForProcess > for the old process, unlike didCommitProvisionalFrame which does. This > appears to be the specific code path where stale entries accumulate -- > the Inspector SI tests like target-cross-origin-page-navigation.html > do main-frame cross-origin navigations that go through > didCommitProvisionalPage. Adding cleanup there would keep > m_instrumentedProcessPageCounts accurate at all times, not just at > teardown. PR #65376 already resolves the receiver-leak assertion via a destructor backstop (~ProxyingNetworkAgent calling removeAllRegisteredReceivers). That fix is sufficient to land. This bug tracks the deeper cleanup: mirror the disableInstrumentationForProcess call from WebPageInspectorController::didCommitProvisionalFrame into the didCommitProvisionalPage path so per-process counts stay accurate during normal navigation lifecycles, not only at WebPageProxy teardown. Repro test: http/tests/site-isolation/inspector/target-cross-origin-page-navigation.html 2213048 1 webkit-bug-importer 2026-05-21 12:29:23 -0700 <rdar://problem/177644735> 2213475 2 bburg 2026-05-22 20:56:10 -0700 Pull request: https://github.com/WebKit/WebKit/pull/65527