GitLab

This is a squashed commit containing bugfixes for the Blockly
update. The individual commit bodies are below as a summary:

1. After using the Paste All Blocks from Backpack menu item, undoing
this operation requires N Undo commands whereas it should only require
1 Undo. This commit wraps the bulk of the pasteBackpack operation in a
try-finally to set/unset the Blockly event group so that all of the
create block events are grouped together.

2. Prevent CSS updates during paste all from backpack

3. Ghica reported that when the backpack is open it is possible to
scroll both the flyout's and workspace's scrollbars and there is no
visual indicator which is which. This means that sometimes the user
will scroll the workspace when intending to scroll the flyout. This
commit makes flyouts hide their parent workspace's scrollbars and show
them after the flyout is dismissed.

4. Improve rendering of flyouts with many blocks

5. Hide backpack with ESC key

6. Only fire parameter orientation event after orientation change

7. The implementation of Blockly.Events.filer on Google's master is
implemented using a O(n^2) algorithm. When performing operations that
affect a large number of blocks, e.g. arrangements or pasting from the
Backpack, this performance is unacceptable. This commit provides a new
implementation of Blockly.Events.filter that is linear in the number
of pending events.

8. Removing mutators for parameters or deleting the procedure block
and then attempting to undo the operation would leave the caller
blocks in an inconsistent state with missing parameters or not
reconnecting the previously attached input blocks. This commit fixes
these issues by changing the order of certain operations so that when
Blockly reverses the order of operations variables, such as the caller
block's PROCNAME, are set correctly.

9. The way that Blockly handles changes to fields breaks how we
renamed parameters in the mutator. This commit provides a custom
handler for key events that makes the renaming behavior match that
before the Blockly update.

10. Cherry-picked code for mutators that fixed the ordering of the
mutator workspace's toolbox with respect to the block canvas.

11. Fixed delete area not appearing for the mutator workspace.

12. Fixed ghost scrollbars from different flyouts capturing the mouse
over event for blocks in the drawer flyout.

13. Made it so that dragging a block in any direction from a flyout
will result in a new block being created. Previously the window was
±70 degrees from the horizontal axis.

14. Improved performance on large projects during load or after moving
a stack of blocks.

15. Fixed error checking on blocks created by typeblock and their
relatives.

16. Fixed null pointer exception when renaming a component prior to
opening the blocks editor.

17. Moved updates to the scrollbar visibility to setScrollbarsVisible
and made all changes to scrollbar visibility go through that method,
which includes fix 16 above.

18. Made arrangement options adjust to workspace scale and
snap-to-grid options.

19. Fixed issue where changing text fields would not result in a
Change event and therefore fail to trigger a workspace save.

20. Added fix to ensure the warning handler exists before attempting
to do error checks on variables.

21. Patched Blockly code to prevent the connection database from
throwing an error when an operation would essentially be an identity
operation on the database.

Change-Id: I0d713e3c92586c0506bfe24d138c634c29d6a208

The buildserver should return an error if Kawa cannot compile the YAIL
file. However, on Windows we end up always pointing at the classes
directory, which always exists, and therefore we continue compiling an
app even though the class files are missing. On Mac/Linux the same
project results in a Kawa compilation error. The logic error is
introduced by the hard coding of the path "/../src/" in
Compiler.java. On Windows the path uses the conventional Windows
separator "\\", so the substring of interest is "\\..\\src\\". This
commit adjusts the lookup string to use File.separator so that the
lookup path will be correct regardless of platform.

Change-Id: Id066cad6710e7fe8c6710b40465c87992f547854

Javac default output format for files is platform-specific. In the
ai.apt macro we do not specify the encoding, so we end up with
ISO-8859-1 strings on Windows that GWT later attempts to decode as
UTF-8 strings. This commit sets the encoding for the javac command to
be UTF-8 so that encodings work correctly on platforms (e.g., Windows)
where the default encoding is not UTF-8.

Change-Id: Iac68786e71983e8605e982a093d3daae09d49c01

Extensions loaded into App Inventor have "More Information" URLs based
on the server's documentation. Obviously, we won't have documentation
for every single extension, nor do we want to update the GAE
application every time someone publishes or updates an extension. This
commit adds a new helpUrl field to the `@DesignerComponent` annotation
that can be used to specify a custom URL for extension documentation.

Closes issue #883.

Change-Id: I7fdc20bb1d5e9cb1bfa257f1ed194748bd5e35ec

For the Internet of Things update, we want to be able to have
collections of related extensions, for example, targetting different
features of the same platform, to be imported as a single unit. This
behavior reflects how we group components such as the LEGO EV3 or NXT
robots and allows us to treat extensions in the same way.

Change-Id: I83e1069421939e57b33689f7ae0a3d5928c6c6cd

For the Internet of Things update, we want to allow extensions to
specify restrictions on designer properties. Specifically, we want IOT
extensions with a BluetoothDevice property to be able to restrict the
component list to the BluetoothLE extension. This commit extends the
behavior of the component property type by allowing an extension
author to include a colon (":") followed by the fully-qualified class
name of the component type to accept.

Change-Id: I6c38890fde99fe947bb1a305412d662694246463

Change-Id: Ieb6fe17306d262ff92c08e28f227767add7005b6

Change-Id: I27191377afac5a2d01db9092934e2c4b25e639ce

The context menu items for arranging blocks and sorting by category
did not have expected behavior due to the new Blockly events
system. This commit introduces two new events StartArrangeBlocks and
EndArrangeBlocks that will manipulate the appropriate flags so the
context menu options work correctly. We also group the arrangement
events in a single group. This makes the undo/redo menu options behave
correctly rather than undoing/redoing individual moves within the
larger arrangement action.

Change-Id: Icb832eef1f91aec5a12025707e9c39b3257987af

Because Blockly is linked into the GWT JavaScript, goog.ui.Dialog
attempts to show itself on the GWT iframe's document rather than on
the top window where the UI is actually displayed. This commit adds a
goog.ui.DomHelper so that the Dialog attachs its DOM elements to the
correct document.

NB: This has the side-effect that the runtime error dialog from the
companion is now UI global (it will show even if in the designer).

Change-Id: I3f7fb60d8898e37a689cbc3f406ef39832717d99

When renaming variables, references to those variables are not
correctly renamed. This is due to a change in the order in which
Blockly manipulates the field's value and calls the validator. Since
the update occurs before validation, we don't know the original name
and therefore cannot find and update all of the references. This
commit provides the flydown with its own implementation of
onHtmlInputChanged_ so that it can provide a behavior that is
compatible with our renaming mechanism, mainly that it will validate
first and only if successful will it update the field's value.

Change-Id: I7569d0f0b1b6f57e476df5e32f1b2a12e7d7533c

Procedure definitions have the option for horizontal/vertical
orientation of parameters. However, switching resulted in the inputs
disappearing from the block. This was caused by the new fields not
being initialized correctly such that when the block attempted to
redraw, expected SVG elements were undefined.

This commit fixes the updateParams_ method for procedure blocks so
that they will correctly handle switching between the two cases.

Change-Id: Ib9a161548954a2990a0351ad07a808c35d2bc928

Creating new procedures that return values in the blocks editor could
result in multiple procedures having the same name. This commit
updates the procedures_defreturn initialization to match that of
procedures_defnoreturn so that renaming is correctly applied.

Change-Id: Ifc30ee27d538b46a61a17d0d4677dad31c4d7d4d

Backpack menu items in the workspace, i.e. Copy All to Backpack and
Paste All From Backpack were broken during the transition to the
latest Blockly. This commit changes any checks for the backpack
presence to use the hasBackpack() test added to WorkspaceSvg.

Change-Id: Iae7919d173f5a5f75003c0132d1f116af9839bd0

Change-Id: I467a9c61a243dc38cd9b311016cf12893b300b66

This commit restores the ability to change the orientation of the
parameters in an event handler block. The functionality is updated to
reflect the new events mechanism in Blockly.

Change-Id: Ib6d9c27b6e13838af886c0abf281e24ba57e6bbe

A change in Chrome results in different behavior when focusing an
invisible element. When the list is longer than the panel, this
results in the panel scrolling-to-top. The fix applied is adapted from
[this issue](https://github.com/gwtproject/gwt/issues/9475#issuecomment-277777989)
on the GWT Github repository.

Change-Id: I8634a95077ee166f34aa265e81a7f272cbd620de

When switching between languages, or loading an AIA created in one
language but the user's language is different, the translation of
event parameters would not always hold. This is due to the use of
`eventparam` as a mutation on the lexical getter and setter
blocks. Through certain actions, the `eventparam` cached value could
be reset to an incorrect state.

This commit does two things to address this issue: (1) adds an
`updateMutation()` method that can be used to recompute the value of
`eventparam` after a blocks update, including changing the dropdown or
attaching a new parent. This keeps the `eventparam` up-to-date during
normal operation (previously it was only computed on save). (2)
FieldParameterFlydown caches the untranslated `eventparam` value for
instantiating new blocks within the workspace. Previously they were
created with the value `null`, which has special meaning as a cached
'not an event' value. Now they are created with the correct mutation
at the beginning and with (1) should maintain the correct value.

Change-Id: I758b8d414d7ddf98f7375ee961187b06b47939cf

This commit brings the App Inventor blocks editor current to the
Google's Blockly. As part of the update, some additional
organizational changes were made to the code:

1. Remove old blockly in lib/ and replaces it with a git submodule
2. Remove old closure-library in lib/ and replaces it with a git
   submodule
3. Move AI2-specific Blockly changes out of core where possible

Change-Id: I58993db05631266d95fbbcc68d49aad1e562fc22

Change-Id: I755616e4c18d77011baaa96dbe852939ee643681

Change-Id: Ie93e0f8f0e738c9f8e4842d5db18d2123119bf6e

maintain backward compatibility for current urls that expect http:// by default

Change-Id: I9c62004bf6b38e06094ebad30f3cd9b789de6510

Change-Id: Icb2747a43848dee97b38b02dfa9cf86bf72ff228

App Engine requires that we now include the companions download
directory in the <static-files> section of appengine-web.xml.

Change-Id: I13c03f0f4625f79ea42a902ab23aaed6adf45569

After the recent App Inventor 2 update, the backpack is now persistent,
which means that when right-clicking on the backpack icon on the block
editor and clicking "Show Backpack documentation", some of the text in
the dialog is incorrect. I have corrected the text to show that of the
release notes.

Change-Id: I223be859193a2b02be8bcf673bdebd6f7116444d

Josh ran into a NullPointerException when receiving a SMS message
using a Pixel running Google Fi. There are a few possible reasons this
could happen:

1. We linked against android.telephony.gms.SmsMessage. If the message
   was formatted as CDMA (3gpp2) then it is possible that the message
   returned is null, resulting in the NPE.

2. Any format other than 3gpp or 3gpp2 will result in the same flow as
   1 if a new format is introduced in the future, or if the format
   were something such as a data packet from Google Voice.

In API 19, there was a new
android.provider.Telephony.Sms.Intents#getMessagesFromIntent
introduced to abstract away extracting SmsMessages from Intents. This
commit uses this API when the SDK is 19 or higher. We also catch NPE
in the broadcast receiver and log it to Logcat to prevent crashing.

Change-Id: I9f00d6156f004d3d72938de1e57b31cfb034d241

We store the backpack in a new “user file” named backpack.xml (the other
user file supported is “android.keystore” for the user’s signing
key). As an optimization we remove the backpack.xml file if we store
“[]” in it (the empty backpack represented as an empty JSON string).

Change-Id: I16fb747f31e59d30535f690c1b35c80bde53629f

When an obfuscated text is used in an application with memory
pressure, the iteration over the obfuscated string can cause a
NullPointerException to be thrown from gnu.text.CharMap. The CharMap
uses weak references to Char objects to make Chars eq?. However, Kawa
does not check whether the Char returned from the WeakReference.get()
method is null and then attempts to take its intValue(). In
applications where there is little memory pressure the WeakReferences
are unlikely to be GCed so we don't normally observe this erroroneous
behavior.

This commit updates kawa-1.11-modified.jar with a check whether the
Char returned by the WeakReference is null prior to taking the
intValue(). If the Char is null, it will fall through to recreate the
Char and add it to the CharMap.

Change-Id: Idb19540b6096a9609a8aa06e77d628cd3ce1dc6b

Limit the number of calls to MockForm.refresh(). This lets a project
with a lot of designer objects load when otherwise the exponential time
and space consumption of repeated and nested calls to refresh() would
cause it to fail to load.

This implementation throttles the calls “locally.” The idea is to limit
the number of calls to 1 every 2 seconds. An alternative implementation
would be to set a flag while a project is loading (which is when this is
a problem) and inhibit all calls to refresh while this flag is set. Then
when the project is no longer loading, clear the flag and do one final
call to refresh. However this would be tricky because loading is
asynchronous and knowing when it is completely finished is
complicated. Such an implementation would be spread over several
modules, and add complexity to the codebase. This change is simpler
because it is limited to one module.

We also add a logging call Ode.CLog to Ode that logs to the developer
console instead of the OdeLog module which logs to the special debug
tab. There are situations where logging to the console is desirable,
like when the problem you are attempting to debug leaves App Inventor
locked up so you cannot see the debug tab!

Change-Id: I5fb9c832704e87c6e73903dc3a75ceb79df5f7f3

On phones with smaller memory footprints, the use of InputStream.mark
was causing an IOException to be thrown that resulted in images not
loading. This patch loads the entirety of the image into a byte[] and
then uses that in conjunction with a ByteArrayInputStream to pass the
image around to the multiple interfaces that need it without reloading
the content. This fixes the failure to load many images in the
emulator.

Change-Id: If7185e8e1a02e19a46037a4dd4df545241fa7f2b

Do not use a transaction when running in the dev server as this causes a
multi-entity group error. This happens because the dev server simulates
the Google Cloud Store using the datastore. If we use a transaction in
create project it winds up covering multiple entity groups. One for our
modifications to the “FileData” kind and a second for the simulated GCS
bucket kind.

Change-Id: I4a31eb16c7866e4ed721c02dbc77addea8df3dbf

Remove the check in versioning.js that throws an exception if a
component version is newer then the system version. This exception
guaranteed that a project touched by a newer version of App Inventor
would not work in the current version.

With this change the project may work if the newer version made not
incompatible changes. The user is shown a warning dialog stating that
the project might not work.

Change-Id: I72c15383b20d918ffd0077d6f7a10f8d1dd4b345

Change-Id: I323066420ea9e0dd9472829e3ee2a1d750d4ba5f

Refactoring to reduce the number of unique dialog boxes instantiated by
Ode. We add this generic MessageDialog method and use it for the
Firebase experimental warning, eliminating the ode.warningDialog
method. Future refactoring can remove more unique dialogs.

Change-Id: I77e1df2a359424aac38c6e8ae8b04fec292cbe99

This reverts most of commit b5e78d70.
We keep the changes to MediaUtil.getBitmapDrawable. So the actual image
fetching happens on a non-UI thread, but the UI Thread blocks until the
network I/O is done. This is non-ideal, as blocking on the UI thread can
result in an Application Not Responding (ANR) error. However this might
be more desirable then having the application apparently not load a
background image for a long time, and then having it suddenly appear
when the background network loading is finally finished.

Change-Id: I3424918a83284d19ee6ccfc9deb8824f5c0605a2

When the minSdkVersion is >= 10 and the user attempts to set an image
to a non-file URL, the Android runtime will throw an
android.os.NetworkOnMainThreadException. This commit changes the
behavior of MediaUtil.getBitmapDrawable to an asynchronous design that
takes a callback that will be triggered after the bitmap has been
loaded. If an exception occurs, an onException callback will be
used. Other failures are reported via the onFailure callback. Due to
the change to AsyncCallbackPair, this change also affects a number of
other components that may or may not have been performing network
access on the main thread.

Testing of this change should focus on the following components:
 * ButtonBase and its descendants
 * Canvas
 * Form
 * GameClient
 * HVArrangement and its descendants
 * Image
 * ImageSprite
 * MediaStore
 * TinyWebDB
 * Voting

Change-Id: Ib9bd0a0a929b26ad44eeda58a7c637398dd70219

Reported by the forum posting [1], the SoundRecorder crashes when the
Start method is called. After some debugging, this is due to
unexpected values being reported via the onInfo method being called
with an unexpected, non-documented value of `what`. The code always
called controller.recorder.stop(), but since the MediaRecorder hadn't
transitioned to the recording state yet, stop() throws an uncaught
IllegalStateException, which kills the program.

This commit makes the onInfo method more robust by reporting known
`what` values via dispatchErrorOccurredEvent and ignoring any values
not declared by AOSP. It also adds a call to StoppedRecording()
otherwise the app may not have been aware of the failure state.

[1] https://groups.google.com/d/msg/mitappinventortest/qMPzn5s82UY/LJu8znfrAwAJ

Change-Id: Ifd12fcf016df07b500b958bd4900781bf6ff6c47

When set this property causes lists to be displayed in JSON notation
rather then Lisp notation. Its default value is “false” so current
projects continue to work and new projects will still use Lisp
format. This can be changed in the future once the Companion that
supports ShowListsAsJson is widely deployed.

Change-Id: I2e6b25c32d27377685eb8111f44536479e7c55b7