There are enough Accounts: messages to separate them from BrowserApp,
and the list is only growing.
This has also the small advantage of removing some non-native event
listeners.
This patch stops referring to package/class objects to identify
Android components directly and instead launches through action intent
filters. The intent filters are scoped to the package, but not marked
as private or as requiring a permission. A malicious package could
inject itself into an account flow, but I don't think there's much
advantage: the only time a secret is passed between activities is when
the native sign up (CreateAccount) and sign in (SignIn) activities
link between themselves, and in this instance I didn't route through
the action intent filters. (This is entirely native -- there's no web
analog -- so I didn't use the indirection.)
When used to do our own animation when opening the browser from the
share overlay. That caused this bug: we didn't call `finish` until
`onAnimationEnd` but since `startActivity` was called, the application
was switched before `onAnimationEnd`, and thus `finish`, could be
called. When we returned to the share overlay, it was in an unexpected
state (`isAnimating` was true) and the user could no longer interact
with it, blocking access to the app the ShareOverlay was opened from.
We fix this by not doing our custom animations and just calling `finish`.
Note: in any case, overriding the animation when opening the browser
could be unintuitive to users because they might expect a consistent
app-switch animation throughout the system.
The call to setResolution has (I believe) not been needed since bug 732971. Prior
to that resolutions used to be applied on the root document in Fennec, and so
browser.js would have to reapply the desired resolution on every tabswitch.
After that bug, the resolution was saved on the content documents for each tab
and so browser.js no longer needed to reapply the resolution. Until recently
doing this was redundant but harmless.
With bug 1180267 though the browser.js code that tracks the resolution may have
the wrong resolution initially, because that is determined in C++ code. Only
after the Java-side code process the setFirstPaintViewport message and sends
that information to browser.js does everything have the correct resolution. In
the case where a tab loaded in the background is brought into the foreground, the
tab-selected code runs before the setFirstPaintViewport code, and therefore uses
an incorrect resolution. This then screws up the viewport clamping code and causes
the page to get scrolled.
The intent of this check is to avoid setting the same margins more than once.
However this is redundant because the code in nsLayoutUtils::SetDisplayPortMargins
already has an equivalent check. Further, this code is wrong because it stores
the old margins per-tab, and so once a new document is loaded the margins may be
the same as "before" but they apply to a different element. In order to be correct
the check would have to track the target element as well as the margin values,
but it's easier to just get rid of this and let nsLayoutUtils handle it.
1) Added some comments to firefox.js to explain the relationship between
extensions.blocklist.interval and security.onecrl.maximum_staleness_in_seconds
2) Modified default values in firefox.js and mobile.js to set maximum staleness
to 1.25x blocklist interval
3) modified the tests_ev_certs.js xpcshell test to cope with larger maximum
staleness values to address test failures
This is a Fennec version of about:accounts, cribbed largely from
Desktop's implementation. This implementation serves two purposes:
One, it allows all fxa-content-server pref handling to remain in
Gecko. Java-side consumers redirect to about:accounts?action=... and
have pref munging and parameter addition (like context=fx_fennec_v1,
etc) handled by about:accounts itself.
Two, it handles network connectivity display and error handling. When
a request is started, we display an animated spinner. We transition
smoothly from the spinner to the iframe display if we can, and if not
we hide any network error and offer to retry. This is more important
in Fennec than it is on Desktop. This approach agrees with Firefox
for iOS.
Some additional notes:
The spinner to iframe transition uses the WebChannel listener to send
LOADED messages to the appropriate XUL <browser> element. It's worth
remembering that Fennec's Gecko is single process, so the <browser> in
question is in the same process. None-the-less, we are close to e10s
safe.
There are four actions: signup/signin/force_reauth, and manage. The
first three try to produce a LOGIN message. The last uses the
fxa-content-server to manage the Account settings. *This is not how
this is arranged on Desktop: Desktop redirects to a new tab, not
wrapped in about:accounts.*
This list is not meant to be exhaustive. There are many ways to get
to the Get Started activity; not all of them are interesting. We're
just trying to capture the significant entrypoints, like firstrun and
preferences.
Pretty straight-forward. It's possible that getStringExtra is not
safe in the face of malicious data; the worst we might expect is a
crash on the Java side; a large memory allocation on the JS side; or
significant URL data transfer. The first is valuable because we'd see
abusers in crash-stats; and the second two are already possible when
opening any URL.
The spinner itself is 60px square, colored like fennec_ui_orange.
The HTML and CSS was hacked out of
https://github.com/lightningtgc/material-refresh, tree
6b1be0046d.
The original code is licensed MIT. I pruned things we don't use,
adjusted the box model for our use, changed the spinner color, and
simplified the CSS.
The default zoom value is only used on the Java side to clamp the min/max zoom
values in the case where zooming is disabled. We can do this much earlier in
the flow, when we are computing the metadata, and reduce the amount of
redundant information being passed around.
From browser.js's point of view there's no difference between restricted and guest profiles. Both use the
parental controls API. So there are only two "simple" solutions here:
* 1) Add a method to nsIParentalControlsService to determine whether the current profiles is a restricted or
a guest profile (Something like isGuest()). But then every platform using this interface would require
to at least implement a stub for this method.
* 2) Add a new restriction that controls installing the theme.
This patch implements option 2. While this restriction is not of much use besides deciding whether we need
to install a specialized theme (DISALLOW_DEFAULT_THEME), it still offers the most flexibility. In a
follow-up bug we could decide to make the restriction configurable by the device admin (requires localized
strings).
This patch creates a copy of the GeckoAppBase style in v21/themes.xml and removes
the custom icons for copy, cut and paste we use on v11+. Instead the system icons,
that match the look&feel of the ActionBar, will be used.
To do this, I ran:
convert <image> -alpha extract -alpha on <image>
The resultant images were slightly larger than their previous
counterparts so I then compressed them with ImageOptim.
DONTBUILD NPOTB
This adds a pass-thru |mach android| command. It's just here to make
it easier to add and remove Android SDK packages: since most folks
don't have the Android tools on their PATH, this saves them having to
root through the object directory to find the path to the tool.
Other 3-dot menu button locations to follow.
I tested that the colors are correct in the various states (e.g. private
browsing).
This patch additionally:
* Restructured the menu button (which allowed the removal of a setVisibility
call
* Removed the now unused tablet assets.
And replace the older assets.
These are the 36dp icons from Google's material design page:
https://www.google.com/design/icons/#ic_more_vert
Then we trim off the whitespace with image magick:
convert icon.png -trim out.png
And compress with ImageOptim.
On Gingerbread devices (Android API 9 or 10) the ViewHelper.setTranslationY code
doesn't work to move the SurfaceView. In order to acheive that effect I turned
LayerView into a ScrollView with a dummy element at the top, and allow it to
scroll. This allows the SurfaceView to move as desired. A few places in the code
were assuming that the LayerView and SurfaceView were always at the same screen
location (which was true before but not any more) and so those sites needed
some updating as well.
The patch removes 455 occurrences of FAIL_ON_WARNINGS from moz.build files, and
adds 78 instances of ALLOW_COMPILER_WARNINGS. About half of those 78 are in
code we control and which should be removable with a little effort.
The annotated methods were using @JNITarget, which would do the same thing.
However, @ReflectionTarget is clearer (like a good variable name).
proguard.cfg states a class with an annotated member will be kept so we
shouldn't need to worry about not being able to retrieve the class via
reflection.
This doesn't appear in Fennec builds and I don't know which web apps "quit"
could appear in so I didn't test this in a web app. I can confirm no issues
with a typical Fennec build, however.
This also removes code to set an icon from an addon, which is unused because,
afaik, addons' icons are also hidden. iirc, there was a bug open on whether we
want addons to be able to show icons (e.g. could be used to show status), but
we can reimplement this if that bug is ever decided.
I was not able to test this with an addon that sets an icon but I did test that
the application did not crash with no addons installed.
On GB devices, ic_menu_new_*tab is in the base menu, unlike the other
configurations, so it stayed on GB.
More complex removals to follow.
I tested this by clicking on all of the base menu items on a Lollipop phone,
KitKat large tablet, and a GB phone.