This is a follow-up to bug 1409249. There are a lot of places where our
factory singleton constructors either don't correctly handle their returned
references being released by the component manager, or do handle it, but in
ways that are not obvious.
This patch handles a few places where we can sometimes wind up with dangling
singleton pointers, adds some explanatory comments and sanity check
assertions, and replaces some uses of manual refcounting with StaticRefPtr and
ClearOnShutdown.
There are still some places where we may wind up with odd behavior if the
first QI for a getService call fails. In those cases, we wind up destroying
the first instance of a service that we create, and re-creating a new one
later.
MozReview-Commit-ID: ANYndvd7aZx
There's a heavy enough overhead to going through XPConnect for
every observer for every visit on the nsINavHistoryObserver
interface, so this patch reduces that by replacing the single-
visit notification with one which accepts an array of visits.
Some notes: To avoid problems with the orderings of the various
ways in which we notify about visits, we have to send our bulk
onVisits notification before doing any of the others. This does
mean it technically behaves slightly different than the prior
approach of interleaving the notifications, but I can't find any
way in which this has any consequences to the end result, and it
doesn't break any tests.
MozReview-Commit-ID: GdeooH8mCkg
Since processing several hundred visits synchronously on the
main thread can be somewhat expensive, and since the main thread
might be idle while the storage worker is running, it makes sense
to chunk the messages into groups of 100.
We did have a concern that this changes the move of the underlying
array to a copy, which might bring some cost with it. For small
arrays, we simply move the underlying array to avoid the copy to
mitigate it. However, for large arrays I was unable to observe any
significant performance cost anyway, so I think we're in the clear.
MozReview-Commit-ID: 1hWSEyKw6pi
During history import, sending NotifyVisited messages from the
chrome process to the content processes in order to change link
colors can take a significant portion of the parent process's
main thread time. Batching it seems to have very significant
results on jank time during history imports.
MozReview-Commit-ID: BHAXpIMa7ly
Because there's an overhead in simply creating/sending/receiving a
runnable, it makes sense to send our onVisits notifications from a
bulk runnable in the case where we're inserting many visits. This
is only step one of many optimizations we can and should make to
the observer system.
MozReview-Commit-ID: Co5yOUCRdnZ
This uses a similar strategy as that employed by moz_places_afterdelete_trigger,
creating a temp table which we write host inserts into, and then deleting all
the rows from it when we're done inserting, effectively resulting in a per-
statement trigger to only do the significant work per host.
MozReview-Commit-ID: 5TUueknq3ng
Right now, NS_GENERIC_FACTORY_SINGLETON_CONSTRUCTOR expects singleton
constructors to return already-addrefed raw pointers, and while it accepts
constructors that return already_AddRefed, most existing don't do so.
Meanwhile, the convention elsewhere is that a raw pointer return value is
owned by the callee, and that the caller needs to addref it if it wants to
keep its own reference to it.
The difference in convention makes it easy to leak (I've definitely caused
more than one shutdown leak this way), so it would be better if we required
the singleton getters to return an explicit already_AddRefed, which would
behave the same for all callers.
This also cleans up several singleton constructors that left a dangling
pointer to their singletons when their initialization methods failed, when
they released their references without clearing their global raw pointers.
MozReview-Commit-ID: 9peyG4pRYcr
I don't bother to label the runnables in the parent process being fired by
VisitedQuery, as we are not planning to perform scheduling in the parent process
if I remember correctly. It would be possible to label those runnables as well.
MozReview-Commit-ID: EosNOu62fEV
Makes initing Places services cheaper, by delaying the connection creation to the first time
it's actually needed.
Same way, delays reading the bookmark roots at the first time they are requested.
Deprecates the concept of lazy observers, since they are no more needed, we can just use addObserver.
Simplifies the startup path: always sends "places-init-complete" (both as a category and a topic) when
the connection starts and adds a "locked" database state when we can't get a working connection.
Makes PlacesCategoriesStarter register for the new category, since it's cheaper than being a bookmarks
observer.
Fixes a couple race conditions in keywords and expiration due to new startup timings.
Removes a test in test_keywords.js that is no more easily feasible, since it'd requires a pre-build
places.sqlite that should be kept up-to-date at every version.
MozReview-Commit-ID: 6ccPUZ651m0
Files which make use of `getPlacesInfo` have been replaced with `History.fetch`.
The code for `GetPlacesInfo` has been deleted from the cpp and idl files.
The test for `getPlacesInfo` has been suitably rewritten and moved alongside the
other History.jsm tests.
There were 2 places where the fact that `getPlacesInfo` takes an array as opposed
to a single uri mattered, in `test_getPlacesInfo.js` and `test_refresh_firefox.py`.
MozReview-Commit-ID: KQSMHCvvlrQ
Root domain icons are no more associated with their pages, BUT if the page uses
a root domain icon from another domain, it should still get an association with it
or we couldn't relate the two.
This also fixes an overlooked problem in PlacesTestUtils where Date objects
cross a boundary and fail instanceof checks. This causes failures in the same
test that this patch is modifying.
To protect from future similar issues some protection has been added to updatedPlaces
so that it will crash in debug builds.
MozReview-Commit-ID: 3MTKhGj3ehj
Root domain icons are no more associated with their pages, BUT if the page uses
a root domain icon from another domain, it should still get an association with it
or we couldn't relate the two.
This also fixes an overlooked problem in PlacesTestUtils where Date objects
cross a boundary and fail instanceof checks. This causes failures in the same
test that this patch is modifying.
To protect from future similar issues some protection has been added to updatedPlaces
so that it will crash in debug builds.
MozReview-Commit-ID: 3MTKhGj3ehj
Additionally, move some history tests to the history folder, split insertMany tests into their own test file.
Also, remove some no more needed android annotations, Firefox for Android doesn't use nor build Places anymore.
MozReview-Commit-ID: 6p4mazeUjsw
The MigrationUtils change is because 99% of the time we will only have
1 visit per URI, and so we spend silly amounts of time doing nothing.
Time spent in composing our undo structure went from ~800ms to ~550ms
with this change.
The other change just seemed obvious - when visits aren't recent,
we shouldn't add them to 'recently visited' lists, which seem to use
'time this function was called' as the time associated with an entry,
which is incorrect.
MozReview-Commit-ID: 2I0D5ApOCI7
This sets a property on the callback object rather than passing an argument
to handleCompletion, to avoid accidentally breaking consumers who don't
expect arguments. The downside is that this required more changes to C++
consumers, but we control all of those so that seemed an acceptable
trade-off.
We should probably actually report errors for all the migrators, but I didn't
want to add risk (what if in some edge-case there are lots?) so I didn't. I'll
file a followup to update them to Cu.reportError() any errors.
MozReview-Commit-ID: Hue9Ci3hyVz
When updating a large number of places, sending runnables to the main thread
for every single one of them whose frecency we update is not conducive to a
responsive UI. This only gets worse once more observers care about these
notifications (e.g. when the library is open).
To avoid this on startup when importing from other browsers, this patch adds
and uses an option to group the frecency notifications. Later patches will
also use the option to avoid other notifications where possible.
MozReview-Commit-ID: D5KqPDu86bo
The new name makes the sense of the condition much clearer. E.g. compare:
NS_WARN_IF_FALSE(!rv.Failed());
with:
NS_WARNING_ASSERTION(!rv.Failed());
The new name also makes it clearer that it only has effect in debug builds,
because that's standard for assertions.
This patch removes checking of all the callback calls in memory reporter
CollectReport() functions, because it's not useful.
The patch also does some associated clean-up.
- Replaces some uses of nsIMemoryReporterCallback with the preferred
nsIHandleReportCallback typedef.
- Replaces aCallback/aCb/aClosure with aHandleRepor/aData for CollectReports()
parameter names, for consistency.
- Adds MOZ_MUST_USE/[must_use] in a few places in nsIMemoryReporter.idl.
- Uses the MOZ_COLLECT_REPORT macro in all suitable places.
Overall the patch reduces code size by ~300 lines and reduces the size of
libxul by about 37 KiB on my Linux64 builds.
This patch makes most Run() declarations in subclasses of nsIRunnable have the
same form: |NS_IMETHOD Run() override|.
As a result of these changes, I had to add |override| to a couple of other
functions to satisfy clang's -Winconsistent-missing-override warning.
