The Java ExoPlayer that we use for HLS support on Android does an accurate
seek, that is, it seeks to the frame at the seek target. This may not be a
keyframe, which we can start decoding at. So change the HLS seek to seek 2
seconds behind the seek target, and drop all frames up to the next keyframe.
This means that after a seek the HLSDemuxer will output a keyframe, and
hopefully (but we can't guarantee of course) it will lie behind the actual seek
target.
We also need to purge the GeckoHlsVideoRenderer's queue of frames which it
is holding onto in order to determine their durations, otherwise after a seek,
we'll get output from this queue of frames. That is, after a seek we would still
get a few frames from the old playback position.
This seek case is particularly problematic as we aggressively shutdown decoders
when the media is paused, including right after the load reaches
loadedmetadata, and we need to seek in order to recover from going dormant.
Differential Revision: https://phabricator.services.mozilla.com/D33254
--HG--
extra : moz-landing-system : lando
In preparation for an <html> root node in browser.xhtml, we need to react to lwtheme changes correctly.
Differential Revision: https://phabricator.services.mozilla.com/D33693
--HG--
extra : source : afa5d3807c8b5810203718e158fa4b185e109f55
Before this commit, we used event telemetry to record timing and counts
for the different steps of a bookmark merge (fetching the local tree,
new local contents, remote tree, and new remote contents; merging;
applying; fetching outgoing records; and notifying observers).
This has several limitations. We need to store all numbers as strings,
include a "flow ID" to tag events from the same merge, and collect
failure reasons twice. We also can't correlate these events to the
existing engine telemetry, meaning we can't see other stats for that
engine, or for the entire sync. Finally, we need to run separate
queries on these events for analysis, instead of extending our
existing engine dashboards.
This approach also feels like an abuse of event telemetry, so this
commit adds a "steps" field for each engine in the Sync ping. Each step
has a name, time taken, and additional named counts, like the number of
items in the tree, or merged structure stats.
Currently, only the `buffered-bookmarks` engine records these steps.
Differential Revision: https://phabricator.services.mozilla.com/D33411
--HG--
extra : moz-landing-system : lando
This commit introduces a `mozISyncedBookmarksMirrorProgressListener`
interface for capturing telemetry and updating shutdown blocker state
after each step of the merge, instead of waiting until the end. This
also means we can also record events for interrupted and failed merges,
and pass validation data through to the Sync ping.
Shutdown hang crash reports now have a `steps` field, indicating the
sequence of completed steps and when they were recorded. If the last
step in the hang report is `fetchLocalTree`, we know the merger is
blocked on `fetchNewLocalContents`. If the last step is
`fetchNewLocalContents`, the merger is stuck at `fetchRemoteTree`,
since that's the next step after `fetchNewLocalContents`.
This commit also implements `Driver::record_telemetry_event` to
dispatch progress callback runnables to the main thread.
Differential Revision: https://phabricator.services.mozilla.com/D31950
--HG--
extra : moz-landing-system : lando
It will cause releasing nsLoadGroup which doesn't like being released off main thread.
Differential Revision: https://phabricator.services.mozilla.com/D33590
--HG--
extra : moz-landing-system : lando
`mozilla.widget.*` was used by old Fennec (Maemo/GTK2). But Android widget doesn't reference this.
### `mozilla.widget.disable-native-theme`
This preference is unused on Android widget. This is for GTK and Windows. Reftests on Android are disabled by `layout/reftests/reftest.list`.
### `mozilla.widget.force-24bpp`
This is unused on Android. This is GTK only.
### `mozilla.widget.use-buffer-pixmap`
No one uses this preference.
Differential Revision: https://phabricator.services.mozilla.com/D33609
--HG--
extra : moz-landing-system : lando
In a branch for `isinstance(obj, (HostSources, HostGeneratedSources))`,
`isinstance(obj, GeneratedSources)` is clearly not going to match
anything. `isinstance(obj, HostGeneratedSources)` is what was intended.
Differential Revision: https://phabricator.services.mozilla.com/D33603
--HG--
extra : moz-landing-system : lando
In preparation for an <html> root node in browser.xhtml, we need to react to lwtheme changes correctly.
Differential Revision: https://phabricator.services.mozilla.com/D33693
--HG--
extra : moz-landing-system : lando
Selecting a different tab is asynchronous, and as such Marionette has
to wait for the "TabSelect" event to be emitted. Further the chrome
window also needs to be focused.
Depends on D33689
Differential Revision: https://phabricator.services.mozilla.com/D33690
--HG--
extra : moz-landing-system : lando
This patch makes several changes to the kinds of URLs where we can cancel
content JS when navigating between them:
1) When navigating directly to a URL (e.g. by typing something into the
location bar and hitting Enter), we allow canceling content JS if the URLs
differ in any way *except* their ref ("#"). To help with this, we also
attempt to fix up the URL (e.g. by prepending "http://" to it).
2) When navigating through history, we allow canceling content JS if the
`prePath` part of the URLs differ. Most notably, this allows canceling
content JS when one of the URLs is an `about:` page (e.g. when hitting the
Home button).
3) We explicitly disallow cancelling content JS if the currently-running JS
is trusted or if the page being navigated away from is anything but
http(s): or file:.
4) We also disallow cancelling content JS for windows that are still being
created (e.g. when creating a new tab or window via `window.open`). For
more background on this, see the comments about `mCreatingWindow` in
dom/ipc/BrowserParent.h.
5) We ensure that, when attempting to cancel JS, the tab ID of the
currently-running script matches the original tab that requested the
cancellation. This avoids a race condition in which a particular JSContext
has already moved on to executing another tab's JS by the time we hit our
interrupt callback.
Differential Revision: https://phabricator.services.mozilla.com/D31875
--HG--
extra : moz-landing-system : lando
The EqualTimeChunker class no longer exists but the test remains. This patch
removes the test for the aforementioned class.
Differential Revision: https://phabricator.services.mozilla.com/D33659
--HG--
extra : moz-landing-system : lando
