This also removes any redundant Ci.nsISupports elements in the interface
lists.
This was done using the following script:
acecb401b7/processors/chromeutils-generateQI.jsm
MozReview-Commit-ID: AIx10P8GpZY
I think that the intermittent error in the bug may be caused by
a pending executeScript call that is somehow handled around the
shutdown of the extension.
To verify this hypothesis, the test now explicitly waits for the
result of the first executeScript call before executing the last
script that is responsible for test completion.
The test should explicitly be checking for the error anyway.
And clean up comments and add reference to bug 1435100 in an
existing comment.
MozReview-Commit-ID: 6gV30Z6zQc4
- Previously, if a port is disconnected by the other end, then memory
would be leaked to `ProxyMessenger.ports` in ExtensionParent.jsm.
To fix this, the port descriptor is now saved separately, keyed by
port ID instead of message manager.
- Previously, when a message manager was disconnected (e.g. window
closed/tab crashed), the port is disconnected only if the port was
created from that page.
This patch adds bookkeeping to keep track of the message managers at
both the sender and receiver's side, so that the port is always
disconnected when the other side goes away.
- The new test browser_ext_port_disconnect_on_crash.js checks whether
the ports are disconnected as expected. Previously, the subtest
connect_from_tab_to_bg_and_crash_tab failed because of the previous
point.
- Although not as deterministic as the crash test, the new
browser_ext_port_disconnect_on_window_close.js reproduces the original
test failure and serves as a regression test for the bug.
- Previously, the data structure in ProxyMessenger.ports contained
the original `sender` and `recipient`. For the purpose of sending
port disconnection messages, these are not necessary and therefore
they have been removed.
- Fix incorrect JSDoc (type of portId is number, not string)
MozReview-Commit-ID: BoaKRVAUKuq
These issues were previously ignored due to the nature of our global import
rules. They need to be fixed before that rule can be updated.
MozReview-Commit-ID: DCChktTc5TW
The old name no longer makes sense, since it no longer exports an spawn_task
symbol, and add_task is what we really care about.
MozReview-Commit-ID: IE7B8Czv8DH
Apparently BrowserTestUtils.removeTab(gBrowser.selectedTab) does not
close the ?discoTest tab, which causes the test to fail eventually.
See comment 6 of bug 1453163 for more details.
MozReview-Commit-ID: 3UgEaVW083i
This patch requests a bit longer test timeout for browser_ext_devtools_panel.js, similarly
to other tests that open/close the devtools toolbox multiple times, because it often
timeouts on linux32 debug builds after we added a couple more "toolbox open/close" cycles
to test Bug 1394750 changes (and also made the devtools theme API tests more readable
by splitting it from the test case that verifies the rest of the devtools panel API).
MozReview-Commit-ID: 3JRWobPRwr5
Use a more specific entry point for the test than
"http://mochitest.test:8888" to make sure that the test is only started
once, when the test opens a tab with the entry point.
MozReview-Commit-ID: 7iAFREDuACu
Extension ports are automatically closed when the message manager of
the source is destroyed. When a tab is detached from a window, its
frameloader is moved to the new window and the original message
manager is destroyed.
Bug 1445537 started listening for SwapDocShells events, but that only
works for the first swap (e.g. detaching a tab once). To avoid early
disconnection of the port, we should continue to subscribe to
SwapDocShells events.
MozReview-Commit-ID: G2ZYAhNyHIL
As we add more behaviors to EventManager, the signature of the constructor
is going to get really clumsy. Head that off by converting it to take a
general parameters object.
This introduces a compatibility problem for existing webextension experiments,
put in a backward-compatibility shim for now.
MozReview-Commit-ID: 72QDfiwRm5j
