GitLab

Previously, AudioNodeOutput::RemoveInput could call AudioHandler::BreakConnectionWithLock
at a time when one of the mutual sets of raw pointers has been updated but the other has
not. This causes the Disable logic to go awry, leading to DCHECK failures and, after
another recent change, ASAN violations, since it can be entered as part of
AudioHandler::DisableOutputsIfNecessary during BreakConnectionWithLock.

This was not obvious by inspection in part because this logic was distributed across
several methods in different classes. To make the order in which these adjustments are
made more clear, factor out the wiring logic from AudioNodeInput and AudioNodeOutput
(and strictly for consistency, AudioParamHandler) into a collection of dedicated
functions.

Call sites are rewritten to call this code, which manages all changes to inputs_,
outputs_ and disabled_outputs_ centrally.

Bug: 910098
Change-Id: Ide367d89cf6601e823169082833d7a41f1019f81
Reviewed-on: https://chromium-review.googlesource.com/c/1359134
Commit-Queue: Jeremy Roman <jbroman@chromium.org>
Reviewed-by: Hongchan Choi <hongchan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#614755}