mirror of
https://github.com/got-feedBack/feedBack-desktop.git
synced 2026-08-10 18:59:55 +00:00
OmikronApexandClaude Fable 5
1359c7c24d
Three identical field crash dumps (2026-07-17, CFG fail-fast 0xc0000409 param 0xa = GUARD_ICALL_CHECK_FAILURE on the JUCE message thread, plugin WndProc frame on the stack) traced to the dispatch timeout desync: every lifecycle call site could give up after 15 s while its op was still queued, leaving two owners mutating engine/plugin lifecycle state with no ordering authority. - New LifecycleExecutor (guide §12 P0): named lifecycle ops, strictly FIFO on the message thread, engine-generation stamped at submit and no-oped when stale. The wait is unbounded with a 15 s watchdog that reports and keeps waiting; false now always means "verifiably did not run" — the "false but it may still run later" state is gone. - initialize/doShutdown bump the generation and run as executor ops; shutdown is the one bounded caller (60 s, then leak the pump — beats hanging process exit behind a wedged driver call). - runDeviceLifecycleOp and closeAllPluginEditorWindows route through the executor; the editor-teardown 15 s give-up (delayed #56 UAF) is removed. - pinPluginModuleForever: plugin modules are pinned on load so JUCE's refcount can never unload them mid-session — a queued window/timer message into an unloaded module is the CFG fail-fast in the dumps. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>