`FlutterDevice.views` is limited by a filter. Pipe this filter up as an
option for the commands that instantiate `FlutterDevice`s. This is the
first change necessary for the CLI tooling to target specific isolates
(#22009).
More work needs to be done after this patch.
* Isolate names are dynamically generated and change every restart.
* This just filters views, not background isolates (`VMService.isolates`).
--track-widget-creation=false to
--track-widget-creation=true
but not when switching from
--track-widget-creation=true
to
--track-widget-creation=false
due to the surprising behavior of Gradle @Optional inputs.
Service extensions can only be activated in debug or profile mode, their code should never be included in release mode. This PR adds guards around all service extension registration calls that enable Dart's tree shaker to remove the extension's code in release mode, which reduces our binary size:
Android Snapshot (uncompressed): minus 127,384 Bytes (-124.40KB)
APK (compressed): minus 38,136 Bytes (-37.24KB)
iOS Snapshot (App.framework, uncompressed): 264,304 Bytes(-258.10KB)
For details: https://docs.google.com/document/d/13JlgvliCn5sWwT2K2SfDwD1NhEfxpJH9DCf22gZZru8/edit
**Benchmark Regressions:** This PR may cause benchmarks to regress because it may change the timing of GC. If you notice a benchmark regression **please note down the exact set of benchmarks that regressed on this PR** and then feel free to revert. I will follow-up with a PR that forces a GC before the effected benchmarks run to get a clean baseline before re-applying this PR.
This attempts to re-land #22656.
There are two changes from the original:
I turned off wrapping completely when not sending output to a terminal. Previously I had defaulted to wrapping at and arbitrary 100 chars in that case, just to keep long messages from being too long, but that turns out the be a bad idea because there are tests that are relying on the specific form of the output. It's also pretty arbitrary, and mostly people sending output to a non-terminal will want unwrapped text.
I found a better way to terminate ANSI color/bold sequences, so that they can be embedded within each other without needed quite as complex a dance with removing redundant sequences.
As part of these changes, I removed the Logger.supportsColor setter so that the one source of truth for color support is in AnsiTerminal.supportsColor.
* Turn on line wrapping again in usage and status messages, adds ANSI color to doctor and analysis messages. (#22656)
This turns on text wrapping for usage messages and status messages. When on a terminal, wraps to the width of the terminal. When writing to a non-terminal, wrap lines at a default column width (currently defined to be 100 chars). If --no-wrap is specified, then no wrapping occurs. If --wrap-column is specified, wraps to that column (if --wrap is on).
Adds ANSI color to the doctor and analysis output on terminals. This is in this PR with the wrapping, since wrapping needs to know how to count visible characters in the presence of ANSI sequences. (This is just one more step towards re-implementing all of Curses for Flutter. :-)) Will not print ANSI sequences when sent to a non-terminal, or of --no-color is specified.
Fixes ANSI color and bold sequences so that they can be combined (bold, colored text), and a small bug in indentation calculation for wrapping.
Since wrapping is now turned on, also removed many redundant '\n's in the code.
We decided that redefining the default for templates was premature. We're going to go back to having "module" in experimental land again, and we'll try again when we have the feature set fully baked.
This keeps the writing of the .metadata files, and writing the template type to them, because that was a good improvement, and there are still a bunch of added tests that improve our coverage.
* Allow passing a restart reason through to analytics
* Update to avoid overlaps with other code
* Remove TODO as this is the real live value
* Improve formatting + constant name
Fixes a gradle error where it was failing to find a plugin because of an absolute path in the .flutter-plugins file instead of a relative path.
I had originally removed this variable because I thought it was redundant with the projectDir, but apparently I was wrong about that (one resolves to a relative path, and one resolves to an absolute path).
This PR reverts that part of the change and reintroduces the (not really) redundant variable.
This turns on text wrapping for usage messages and status messages. When on a terminal, wraps to the width of the terminal. When writing to a non-terminal, wrap lines at a default column width (currently defined to be 100 chars). If --no-wrap is specified, then no wrapping occurs. If --wrap-column is specified, wraps to that column (if --wrap is on).
Adds ANSI color to the doctor and analysis output on terminals. This is in this PR with the wrapping, since wrapping needs to know how to count visible characters in the presence of ANSI sequences. (This is just one more step towards re-implementing all of Curses for Flutter. :-)) Will not print ANSI sequences when sent to a non-terminal, or of --no-color is specified.
Fixes ANSI color and bold sequences so that they can be combined (bold, colored text), and a small bug in indentation calculation for wrapping.
Since wrapping is now turned on, also removed many redundant '\n's in the code.
This all happened because I was trying to be a little too helpful...
Part of the job of the "create" command is to recreate missing pieces of existing projects, and now that the default has changed, I wanted to make it so that if someone had created a default flutter create project before, that they could run a default flutter create there again, and not have it trashed by using the new default template (application) over the old one (app).
This meant I had to detect what type of project it was. Unfortunately, in the past we didn't write anything in the .metadata file to identify the type of project, and since the goal was regenerating missing files, I can't count on anything existing, so it's just a heuristic match.
This simplifies the heuristics down to just detecting the difference between "app" and "application" projects, and only detect the other types if they're explicitly listed in the .metadata file (I changed the code in my original PR to add the project type to the .metadata file). People used to have to specify the type for those anyhow, so it shouldn't be a surprise to users.
So, the main difference in the new heuristics from my last attempt is that if you have a directory that has some other stuff it (like maybe a "plugin" project), then we'll recreate (pronounced "mess up") the project using the "application" template, but that was true before (except it would use the "app" template).
Fixes#22726
This renames the "module" template to the "application" template, and makes "application" the default. The existing "app" template is now deprecated.
flutter create also now recognizes the type of project in an existing directory, and is able to recreate it without having the template type explicitly specified (although you can still do that). It does this now by first looking in the .metadata file for the new project_type field, and if it doesn't find that, then it looks at the directory structure. Also, the .metadata file is now overwritten even on an existing directory so that 1) the project_type can be added to legacy projects, and 2) the version of Flutter that updated the project last is updated.
I also cleaned up a bunch of things in create_test.dart, added many more tests, and added an example test to the test/ directory in the generated output of the application template.
Fixes#22530Fixes#22344
Previously flutter_tools had used "gradle properties" to find the build types
and flavors supported by the Gradle project. Tasks should work more reliably
across different versions of the Android Gradle plugin.
Fixes https://github.com/flutter/flutter/issues/20781
* Change file paths to URIs to fix "Could not run configuration in engine" on Windows
Plus unskip test that was failing due to this.
Fixes#21348.
* Remove unused import
The asset subsystem uses URIs to represent asset locations. The font manifest
should also use URI-encoded paths instead of unencoded paths taken directly
from the manifest YAML.
Fixes https://github.com/flutter/flutter/issues/19452
Eliminates the --snapshot and --depfile parameters from the flutter
bundle command. The snapshot parameter is unused in Dart 2 -- code is
built to kernel .dill files and for profile/release builds, then AOT
compiled.
While depfiles are still used in Dart 2 (e.g. by the kernel compiler),
there are enough assumptions in the code that they lie in the default
location (e.g. in the Gradle build) and no reasons to support
user-cusomisation that it makes sense to eliminate the --depfile option
as well, and always use the default location.
This commit also renames 'depFilePath' to 'depfilePath' for consistency
across the codebase.
This also involves switching from Core JIT to App JIT snapshot, and replacing per-isolate VM snapshot with the shared VM snapshot.
For now there is no separate update bundle file, as the generated update gets packaged directly into the APK for testing purposes.
* Fix xcode_backend.sh script to support add2app
* Fix ios deployment target. Too old for new Xcode.
* Fix ios host app
* Register plugins with Flutter view
* Prototype
* Fix paths to Flutter library resources
* Invoke pod install as necessary for materialized modules
* Add devicelab test for module use on iOS
* Remove debug output
* Rebase, reame materialize editable
* Add devicelab test editable iOS host app
* Removed add2app test section
Swap out the moon emoji used for progress spinner for a single-cell character.
The moon emoji looked cool, but couldn't be used because of bugs in xterm.js, used for VSCode's terminal, among others. The moon emoji is two character cells wide, but xterm.js doesn't advance by two cells when it adds the emoji, but does go back by two when it backspaces.
This changes us to a different character animation (dots) that is only one cell wide, and so doesn't have this problem.
* Improve documentation and clean up code.
* Remove "Note that".
The phrase "note that" is basically meaningless as a prefix to an
otherwise fine sentence.
This changes the compiler output for gradle to be less verbose and more easily read.
This only applies to compilation error messages: other gradle messages will continue to print as before.
It also fixes a small problem with the performance measurement printing (see that "7.1s" on it's own line in the original?) so that if something is expected to have multiple lines of output, it prints an initial line, and a "Done" line with the elapsed time, so that it's possible to know what the time applies to.
It also updates the spinner to be fancier, at least on platforms other than Windows (which is missing a lot of symbols in its console font).
Addresses #17307
This adds support to AnsiTerminal for colored output, and makes all tool output written to stderr (with the printError function) colored red.
No color codes are sent if the terminal doesn't support color (or isn't a terminal).
Also makes "progress" output print the elapsed time when not connected to a terminal, so that redirected output and terminal output match (redirected output doesn't print the spinner, however).
Addresses #17307
Xcode 10 introduces a new build system which includes stricter checks on
duplicate build outputs.
When plugins are in use, there are two competing build actions that copy
Flutter.framework into the build application Frameworks directory:
1. The Embed Frameworks build phase for the Runner project
2. The [CP] Embed Pods Frameworks build phase that pod install creates
in the project.
Item (1) is there to ensure the framework is copied into the built app
in the case where there are no plugins (and therefore no CocoaPods
integration in the Xcode project). Item (2) is there because Flutter's
podspec declares Flutter.framework as a vended_framework, and CocoaPods
automatically adds a copy step for each such vended_framework in the
transitive closure of CocoaPods dependencies.
As an immediate fix, we opt back into the build system used by Xcode 9
and earlier. Longer term, we need to update our templates and
flutter_tools to correctly handle this situation.
See: https://github.com/flutter/flutter/issues/20685
This caused issues for projects without an Xcode workspace. Almost all
Flutter projects in the wild will have a workspace, but this patch needs
to add a check to catch any that lack one.
This reverts commit 021f472efc.
Xcode 10 introduces a new build system which includes stricter checks on
duplicate build outputs.
When plugins are in use, there are two competing build actions that copy
Flutter.framework into the build application Frameworks directory:
1. The Embed Frameworks build phase for the Runner project
2. The [CP] Embed Pods Frameworks build phase that pod install creates
in the project.
Item (1) is there to ensure the framework is copied into the built app
in the case where there are no plugins (and therefore no CocoaPods
integration in the Xcode project). Item (2) is there because Flutter's
podspec declares Flutter.framework as a vended_framework, and CocoaPods
automatically adds a copy step for each such vended_framework in the
transitive closure of CocoaPods dependencies.
As an immediate fix, we opt back into the build system used by Xcode 9
and earlier. Longer term, we need to update our templates and
flutter_tools to correctly handle this situation.
See: https://github.com/flutter/flutter/issues/20685
This is a blocker for Google roll since we are not at dev4.0:
- Future is not yet part of dart:core.
- Future.sync().then<dynamic>... causes failure without the new keyword.
* Don't set the `vmServices` member variable until it's fully initialized.
* Add a timeout to the future that sends the 'started' event to the IDE
https://github.com/flutter/flutter/issues/16604
Now that Dart 1 support has been eliminated, generator should always be
populated to a kernel compiler instance.
Also moves the constructor to the top of the class, and orders final
fields before non-final fields, as dictated by the style guide.
This tickled a bug in KernelCompiler.compile() where the fingerprinter
doesn't include the outputFilePath in its list of dependencies. As such,
if the output .dill file is missing or corrupted, the fingerprint still
matches and re-compile is skipped, even though it shouldn't be. I'll fix
that in a followup, then look at how this triggered that issue. My
hypothesis is that that it's due to the aot kernel compile and bundle
kernel compile have separate output directories for the .dill files
(build/ vs build/aot) but the same output directory for the associated
depfiles (due to this patch).
This reverts commit 43a106e95a.
The --snapshot argument was only necessary in Dart 1. The --depfile
argument was only used in Dart 2 mode to pass to the kernel compiler,
but was inconsistent with the 'build aot' command, where the depfile was
always set to build/kernel_compile.d.
This patch updates 'build bundle' to emit the depfile to a location
consistent with the 'build aot' command; since it's not intended to be
user-configurable and flutter.gradle hardcodes the location to
build/kernel_compile.d either way, this patch also eliminates the
ability to configure the filename altogether.
* Add a detach command to detach without terminating (#21376)
* Add a detach command to detach without terminating
Fixes#21154.
* Bump protocol version for app.detach
* Tweak to detach/quit text
* Change logPrefix to named param
* Fix the text that the devicelab attach test looks for
* Allow FlutterTester to be provided with the working directory for execution
Previously this test set fs.currentDirectory which prevents running tests concurrently. This allows setting the working directory for a FlutterTester in the cosntructor (optionally) and passes it through from the test (without setting fs.currentDirectory).
* Remove trailing whitespace
* Add a detach command to detach without terminating
Fixes#21154.
* Bump protocol version for app.detach
* Tweak to detach/quit text
* Change logPrefix to named param
* First step in Flutter Doctor refactor. Assigns categories to all validators.
* Revert "Roll engine e54bc4ea1832..a84b210b3d26 (6 commits) (#20453)"
This reverts commit 05c2880a17.
* Split iOS and Android workflows into workflow and validator classes.
* Change ValidatorCategory to handle standalone validators that share a
category (e.g. IntelliJ).
Also make Android Studio and Android toolchain use separate categories.
At this stage, flutter doctor output matches what it was previously.
(The summary() method itself has not yet been changed )
* Change doctor summary code to support validator categories.
Output is still unchanged.
* Handle small formatting issues.
* Flip Flutter category's isGroup field to false until it's actually
needed.
* Revert auto-generated formatting changes to keep those lines from
muddying the pull.
* Small fixes pointed out by analyzer.
* Properly fix analyzer issues around const constructors.
* Small changes to address comments.
* Add tests to verify grouped validator behavior and validationtype
merging.
* Update doctor.dart
* Add comments for clarification.
This patch eliminates the --preview-dart-2/--no-preview-dart-2 flag,
hardcoding all uses to true. It also defaults all previewDart2 method
parameters to true, where they hadn't yet been.
A series of subsequent patches will eliminate all previewDart2
parameters and the associated code from within the codebase.
DevFS.update only runs in Dart 2 mode when the generator parameter is
supplied. In Dart 2 mode, both mainPath and pathToReload are required
parameters; this patch marks them as such.
generator is required for running in Dart 2. All call sites other than tests already explicitly set this value.
Note the statements on line 510 and line 516 for why mainPath and pathToReload are required.
Its type uses a generic type which is limited to library-visibility.
Eliminating compilationQueue from ResidentCompiler's public interface
makes it possible to mock in tests.
When switching between channels, we were leaving around the version freshness stamp file (bin/cache/flutter_version_check.stamp), which meant that the flutter tool would read from that file to see what the cached date of the most recent commit to the current channel (branch) was. The problem was that since the file was created while on the previous channel, the cached date was for the wrong channel, so if you switch from master to beta, flutter would think that the channel was out of date, and a new version was available, at least for three days after the first time it checked (after three days since the last time the freshness was checked, the cached date would get updated).
This PR modifies the channel command to remove that stamp file whenever the user switches channels, so that the cached date will be from the right channel when it is recreated.
Fixes#21134
* Fix coverage collection crash
Based on Jason's patch in https://github.com/flutter/flutter/pull/19546/
This is more or less the same but I tried to avoid using `dynamic`.
* Improve argument and variable names in flutter_platform
* Don't bother with reduce, since the order is guaranteed.
* Start logger in startProgress to avoid assertion failure
There are lots of places that create a Status() and all but two of them call ..start() immediately. This is one of the places that doesn't, which causes an assertion failure when running with --enable-asserts and the other is in the same file (I suspect it's also incorrect, but possibly it's never used - I'll look at that separately when I can trace some code that calls it).
Fixes#20812.
* Enable asserts for tools tests
* Fix lint
* Rename enableAsserts -> enableFlutterToolAsserts
To make it clearer that it only enables asserts for flutter_tools when set.
flutter_tools cannot depend on flutter_goldens (as flutter_goldens
depdends on the Flutter sdk), so this commit splits client.dart from
flutter_goldens to a pure-dart flutter_goldens_client package.
* `flutter analyze` cleanup
* Make `--dartdocs` work in all modes.
* Make `analyze-sample-code.dart` more resilient.
* Add a test for `analyze-sample-code.dart`.
* Minor cleanup in related code and files.
* Apply review comments
* Fix tests
* Fixes resulting from audit of issues links
I looked at every link to GitHub in our repo. For cases where we had a TODO that was waiting for a bug to be fixed, and the bug has now been fixed, I applied the pending change. For cases where the link was out of date, I updated the link.
* Update run_test.dart
skip this test again since it failed on linux and macos bots
All temporary directory start with `flutter_` and have their random component separated from the name by a period, as in `flutter_test_bundle.YFYQMY`.
I've tried to find some of the places where we didn't cleanly delete temporary directories, too. This greatly reduces, though it does not entirely eliminate, the directories we leave behind when running tests, especially `flutter_tools` tests.
While I was at it I standardized on `tempDir` as the variable name for temporary directories, since it was the most common, removing occurrences of `temp` and `tmp`, among others.
Also I factored out some common code that used to catch exceptions that happen on Windows, and made more places use that pattern.
* Try to resolve an intermitted crash during coverage collection
The only theory I can come up with is that maybe the test completes
before we finish processing the standard input, so I made the test
harness wait for the observatory URL before considering whether the
test has finished or not.
Also, some code cleanup while I'm at it, e.g. avoiding using "onFoo"
for the names of methods, avoiding back-to-back switch statements with
the same values, avoiding `_` argument names, and using `?.` instead
of `if (foo != null) foo.`.
* Revert back the signature of _pipeStandardStreamsToConsole
* Also remove the other additions to this method.
* Reland "Roll engine to version b148e628ec86b3a9a0382e0bcfae73f0390a8232 (#20427)"
This is a re-land with downgraded `package:flutter_gallery_assets`
version.
* Downgrade package:flutter_gallery_assets to 0.1.4
* Change engine.version to 81baff97c29bb08cbf8453a3f9042c5813f84ad3 (which contains an additional fix)
* Change engine.version to e3687f70c7ece72000b32ee1b3c02755ba5361ac (since mac tarballs are corrupted on earlier commit)
Reason for revert: The package:flutter_gallery_assets has removed some images which are required for the examples/flutter_gallery, so the gallery build is failing (only discovered after landing, since gallery doesn't seem to get built during github PR presubmit checks)
This CL
* rolls `engine.version` to flutter/engine@b148e628 (which includes dart sdk 2.1.0-dev)
* rolls `goldens.version` to flutter/goldens@6c45fafdf (which includes updates due to skia changes in engine)
* changes `platform.dill` to `platform_strong.dill` in various places due to flutter/engine@a84b210b
* adds explicit `environment: sdk: ">=2.0.0-dev.68 < 3.0.0"` constraints to `pubspec.yaml` and `pubspec.yaml.tmpl` files (since pub defaults to `<2.0.0` if omitted)
* upgrades to newer versions of various 3rd party packages (to ensure transitive dependencies have `<3.0.0` sdk constraint)
* Upgrade everything except matcher.
* Roll matcher (and test)
* Adjust tests that depend on flutter:test directly to depend on a shim
* Require use of package:test shim and remove other references to package:test
In certain cases, the test would fail before creating the (lazily created) compiler object, and then we'd
try to call shutdown() on null in those cases.
Fixes#18610
This removes the final traces of Travis and Appveyor from the Flutter tree.
I've updated the documentation and fixed a couple of places where scripts look for Travis, and eliminated the dart tools runningOnTravis function (which was unused anyhow).
There are places in the flutter script that used to look for the environment variable TRAVIS. We actually do want to continue to detect that we're running on Travis there, since in the plugins repo we still use Travis (for the moment). In any case, it's OK, because the CI environment variable is set on all of the CI bots (Cirrus, Travis, and Appveyor).
FastLane doesn't have a setup_cirrus equivalent to setup_travis, but it actually doesn't matter there either, since it doesn't do Travis-specific things, and it also looks for the CI environment variable.
//mobile/flutter/tests/app:basic_runner_test_*
during a google3 roll.
The following exception was being thrown
_TypeError: type 'String' is not a subtype of type 'File'
```
8)
<asynchronous suspension>
utter_command.dart:347:18)
<asynchronous suspension>
/flutter_command.dart:282:33)
<asynchronous suspension>
xt.dart:142:29)
<asynchronous suspension>
Disallow calling stop() or cancel() multiple times. This means that
when you use startProgress you have to more carefully think about what
exactly is going on.
Properly cancel startProgress in non-ANSI situations, so that
back-to-back startProgress calls all render to the console.
Disallow calling stop() or cancel() multiple times. This means that
when you use startProgress you have to more carefully think about what
exactly is going on.
Properly cancel startProgress in non-ANSI situations, so that
back-to-back startProgress calls all render to the console.
* Add `--dry-run` option to `flutter format` sub command
* Add `--set-exit-if-changed` option to `format` sub command
* Add `--machine` option to `format` sub command
* Make variable names to be not shorthand: cmd -> command
Fix https://github.com/flutter/flutter/pull/18360#discussion_r199656120
We don't actually care if we're on linux, we care only if the lcov
tool is available. We check for that explicitly just below.
Additionally there is code below which indicates that lcov is available
for macOS. Before this change that code would never execute.
* Extract some of startApp into a reusable method
* Get basic attach --machine working
* Attach --machine tweaks
Move validation to validate method and create daemon early so we get the startup event before trying to get a connection.
* Bump daemon version so we know whether it's valid to flutter attach
* Tweak output text
* Swap package imports for relative
* Review tweaks (naming, formatting, typedefs)
* Separate arguments from process spawning
This will make calling attach easier
* Add a basic test for flutter attach --machine
* Fix crash if port unforward modifies the list of forwarded ports
* Add a no-op port forwarder for flutter-tester
* Switch to using BasicProject instead of our own inline code
* Fix expectation in test now we have a portForwarder
* Remove stale TODO (this is done)
* Tweak formatting
* Change some Completers to void to fix Dart 2 issues
* Make cleanup of flutter processes in tests more reliable
* Fix quit signals
I confused SigInt&SigTerm for SigTerm&SigKill when I first did this. SigTerm can be blocked and doesn't guarantee the process will be terminated.
* Don't use deprecated constants
* Fix typo
* Add some additional info to debug buffer
* Fix return types on Futures
* Only printError on our simple messages
Any other type is a real error that should be a normal crash (to get a proper error log).
See #19453.
* Add toList() to convert Iterable<String> -> List<String>
This code previously throw in Dart-2 mode.
Fixes#19453.
* Move getSimulatorPath into Xcode
* Add a test that we tried to launch the simulator
* Remove unused import
This uses @kevmoo's completion package to do command line completion for flutter, and a new command "bash-completion" (with alias "zsh-completion") that will output the necessary shell script setup code, and adds the hidden command "completion" that does the actual completion.
Because it adds a dependency, I also had to do flutter update-packages --force-upgrade.
Fixes#18988.
* Revert "Revert "Roll engine to fed2ea458ed49088d33eddabc546ba56d600c717 (includes dart roll) (#19044)" (#19276)"
This reverts commit cf932490b7 as it also
includes fix for type error that broke tests.
* Add type cast for dart2 type checks.
* Move up to latest goldens
* Make inDirectory() type-parameterized.
* Add typecasting to transitions_perf_test.dart and microbenchmarks.
* Add boolean flag initialization in save_catalog_screenshots.dart
* Add type conversion to gallery transition test
ios-deploy relies on LLDB.framework, which relies on /usr/bin/python and
the 'six' module that's installed on the system. However, it appears to
use the first version of Python on PATH, rather than explicitly
specifying the system install. If a user has a custom install of Python
(e.g., via Homebrew or MacPorts) ahead of the system Python on their
PATH, LLDB.framework will pick up that version instead. If the user
hasn't installed the 'six' module, ios-deploy will fail with a
relatively cryptic error message.
This patch pushes /usr/bin to the front of PATH for the duration of the
ios-deploy run to avoid this scenario.
This patch also removes checks for package six.
Neither Flutter nor any of its direct dependencies/tooling relies on
package six. ios-deploy depends on LLDB.framework (included with Xcode),
which relies on a Python script that imports this package but uses
whichever Python is at the front of the path. Flutter now invokes
ios-deploy with a PATH with /usr/bin forced to the front in order to
avoid this problem.
We could have retained the check out of paranoia, but this seems
unnecessary since it's entirely possible LLDB.framework may one day drop
this dependency, in which case I'd expect the base system install of
Python would likely drop it as well.
* Roll engine to rolled dart
Add connectionTimeout property to class that extends HttpOverrides.
Upgrade packages.
Add more type fixes.
Update goldens hash.
Add more type fixes.
Upgrade packages.
Fix tests.
Remove strong-mode from analysis options yaml file.
Increase dev/devicelab timeout from 2 to 10 seconds.
Added getVM to waitForViews. Fix type errors in compileExpression.
* Pick up engine revision with fix for microbenchmarks regression.
Increase default timeout from 20/30 to 60s.
They should fail, because we don't have real support yet.
Also, make the debug override work in release builds, so that people
on those platforms have a workaround.
This will let us re-use that code from fuchsia_tester.dart.
Tested by running the stocks example tests with coverage collection before and after.
.lcov files are identical.
* Adding a setup function for hot restart that can be overwritten to perform any necessary initialization before a restart is started.
* Adding a test for setupHotRestart
* Removing local pubspec change
* Fixing some type annotation errors
* Responding to comments
* Remove the 'app' domain from flutter daemon
By default the daemon won't register the "app" domain, you need to opt-in (which the 'run' command does, as well as the tests for the app functionality).
Fixes#6658.
* Tweak text
* Put restart/callServiceExtension/stop back into daemon mode
* Add a comment about removing discoverApps
* Add --create option to flutter emulators
* Tweaks to error message
* Simplify emulator search logic
* Make name optional
* Add a note about this option being used with --create
* Tweaks to help information
* Switch to processManager for easier testing
* Don't crash on missing files or missing properties in Android Emulator
* Move name suffixing into emulator manager
This allows it to be tested in the EmulatorManager tests and also used by daemon later if desired.
* Pass the context's android SDK through so it can be mocked by tests
* Misc fixes
* Add tests around emulator creation
Process calls are mocked to avoid needing a real SDK (and to be fast). Full integration tests may be useful, but may require ensuring all build environments etc. are set up correctly.
* Simplify avdManagerPath
Previous changes were to emulatorPath!
* Fix lint errors
* Fix incorrect file exgtension for Windows
* Fix an issue where no system images would crash
reduce throws on an empty collection.
* Fix "null" appearing in error messages
The name we attempted to use will now always be returned, even in the case of failure.
* Add additional info to missing-system-image failure message
On Windows after installing Andriod Studio I didn't have any of these and got this message. Installing with sdkmanager fixed the issue.
* Fix thrown errors
runResult had a toString() but we moved to ProcessResult when switching to ProcessManager to this ended up throwing "Instance of ProcessResult".
* Fix package import
* Fix more package imports
* Move mock implementation into Mock class
There seemed to be issues using Lists in args with Mockito that I couldn't figure out (docs say to use typed() but I couldn't make this compile with these lists still)..
* Rename method that's ambigious now we have create
* Handle where there's no avd path
* Add another toList() :(
* Remove comment that was rewritten
* Fix forbidden import
* Make optional arg more obviously optional
* Reformat doc
* Note that we create a pixel device in help text
* Make this a named arg
* Improve update checking
This change emables pinging the server to check for updates regardless of whether the local version is "out of date". The server code already has a 7-day cache so the result is that we can now ping the server once every 7 days instead of waiting for the local install to be 4 weeks out of date before pinging.
The original 4 week period is still used for when we'll start warning the user they're out of date if we could not confirm with the server whether there's a new version.
* Improve message when we know there's a new version available
* Fix bnullable bool checks
* Switch nullable bool to enum
* Fix casing of enum values
* Remove stale comment
The names are now descriptive so doesn't need additional explanation.
* Improve name of function
* Remove note:
* Rename kPauseToLetUserReadTheMessage -> timeToPauseToLetUserReadTheMessage
* Change kVersionAgeConsideredUpToDate to 5 weeks from 4
* Inline the isNewerFrameworkVersionAvailable check
* Fix indenting (?)
* Fix more indenting
* Rename function to be clearer it's getting the commit date
* Formating tweaks
* Update stamp when connection failed, and reduce time before we'll try again
Previously we would hit the server on every command if we thought we might be out of date and we never successfully connected (eg. if you're offline). This makes the stamp update even when there's a conneciton failure so that this won't happen, but reduces the time till we check again from 7 days to 3 days to compensate a little in case it was a one-off.
https://github.com/flutter/flutter/pull/18193#issuecomment-399222269
* Fix comment
* Don't perform update checks if not on an official channel
* Revert changes to core SDK members that haven't been propagated to the g3 Dart SDK yet.
* Respond to comments.
* Remove retype
* Temporarily un-remove VMService.retype to unblock g3 roll
* Checkout back to master, we need fewer changes to reach compat with dev.61 than dev.60
Seems like VS Code may have started using lowercase extension folders recently (our identifier is `Dart-Code.flutter` but now the folder is named `dart-code.flutter`), so this makes the check not sensitive to casing.
Also reuse extension identifier for download link
fuchsia_tester.dart still assumes Dart 1. Previously, it ran tests directly
from source, flutter_platform.dart automatically runs a kernel compile when
operating in Dart 2 mode, but this assumes a functional Dart SDK is available
in the artifacts directly, and fuchsia_tester.dart mocks out the artifacts
directory with an empty temp dir.
Remaining work is:
1. Get the frontend server building as a dependency on Fuchsia.
2. Patch fuchsia_tester.dart to use a valid Dart SDK and frontend server.
This also reverts migration to Dart 2 typedef syntax.
This reverts commit 6c56bb2. (#18362)
This reverts commit 3daebd0. (#18316)
Now that Dart 1 is turned off, reapplying my change to turn on the prefer_generic_function_type_aliases analysis option, and fix all the typedefs to Dart 2 preferred syntax.
Also eliminated the unused analysis_options_repo.yaml file and turned on public_member_api_docs in analysys_options.yaml.
No logic changes, just changing the typedef syntax for all typedefs, and updating analysis options.
* It's time to #deleteDart1 (#18293)
Eliminates support for Dart 1 in flutter_tools, and drops our Dart 1
benchmarks. All commands now run in Dart 1 mode only.
Eliminates --preview-dart-2 / --no-preview-dart-2 support.
* Fix indentation, remove no longer necessary .toList()
* Only push udpated kernel if >0 invalidated srcs
Eliminates support for Dart 1 in flutter_tools, and drops our Dart 1
benchmarks. All commands now run in Dart 1 mode only.
Eliminates --preview-dart-2 / --no-preview-dart-2 support.
DevFS.update() returns the number of bytes synced from the host to the
target device. When running in Dart 2 mode, ensure that the size of the
kernel file is included in the count of bytes synced.
Iterates over a copy of the dirty entries list rather than the original,
in order to avoid the potential for a concurrent modification exception
when fileUris are removed/added later in the loop.
This occurs when non-asset URIs (Dart sources) are listed in the dirty
set when --preview-dart-2 is enabled (the default).
Now that we have thousands of tests, it doesn't make sense to display a separate line for each test. The result is just megabytes of logs that you have to scrub through to find error messages.
We were already skipping the check in `flutter upgrade` but this also spawns `flutter precache` and `flutter doctor` which would still check and potentially output the message.
Uses the `version` property from the `pubspec.yaml` file to set the corresponding fields in the `local.properties` file respectively in the `Generated.xcconfig` file.
The `--build-name` and `--build-number` options have changed. Now they trump the `version` property from the `pubspec.yaml` file.
If the `version` property is not set and the `--build-name` and `--build-number` options are not provided, the build command will not change the `local.properties` / `Generated.xcconfig` file.
* Send an event at startup with the protocol version and pid
The pid will help with some of the issues of terminate the process when launched through a shell script and the version will allow clients to make decisions about supported features.
I've also bumped the protocol version number for two reasons:
1. This change
2. We didn't increase it when we added the previous emulator commands
When running with --local-engine, build single-architecture apps, rather
than what's declared in the Xcode project.
By default, Flutter creates Xcode projects that generate universal
binaries supporting both armv7 and arm64 processors. Since local engine builds
are all single-architecture, override the Xcode project's ARCHS setting
with the appropriate value for the specific engine.
NOTE: this assumes engine paths consistent with those used in the engine
repo: 32-bit iOS engine builds end in _arm, 64-bit builds do not.
Our style guide says the k's are not necessary, and it seems like a good idea to make all the code be consistent on this.
Only naming changes to private vars: no logic changes.
Further digging revealed that the reason --no-sim-use-hardfp was
required to be specified explicitly was that Android engine gen_snapshot
binaries are built on Windows with target_os=win.
Previously, the only hook into the rasterizer schreenshot code path
was via FlutterDriver.screenshot. This adds the ability for the
Flutter tool to hook into all three types of screenshots:
1. device (e.g. `adb shell screencap`)
2. Skia (capture spk)
3. Rasterizer (capture PNG via the engine)
This reverts #17147, which was safe under the assumption that
getArtifactPath() returns a platform-specific gen_snapshot instance
(which have the correct default set for this flag) -- it turns out that
though we pass the platform to getArtifactPath(), we always return the
host gen_snapshot for Android.
A followup patch will update getArtifactPath and revert this patch.
We've always cached kernel compiles during bundle builds. This adds
caching for kernel compiles during AOT builds by moving caching into the
KernelCompiler.compile() method rather than around each invocation of it.
This also filters buildbot paths included by the kernel compile that are
not present on the local machine at paths /b/build/slave/Linux_Engine/...
The kernel compiler should probably include an option to not emit these
paths in the depfile, since these are used both by Gradle and the
Fingerprinter class.
Previously, Flutter did not support iOS devices with armv7 or armv7s
CPUs. We now support these devices. This eliminates the previous
hardcoded checks that prevented running on simulators of older devices.
We maintain the existing restriction on running on watchOS or tvOS
simulators.
Previously, in non-release (i.e. profile) AOT builds, we were setting
--no-checked and --conditional_directives flags. --no-checked is the
default, and we don't make use of conditional directives in Flutter.
Allows users of Fingerprinter to filter the set of paths collected from
the explicitly-specified paths and those collected from depfiles.
In some cases, depfiles are emitted with files that are not present on
the local disk (e.g. the frontend compiler currently emits buildbot
paths for the dart core libraries and dart:ui). These files will not
materially affect whether we need to re-run a build action for which
they are inputs, since they're not present in the filesystem and
therefore cannot change.
Depending on whether can read the AVD (or it even has all fields populated) we might get extra "empty" columns, so this trims all blank cells from the end.
Emaultor keeps running on a seuccessful launch, so this automatically returns after 3 seconds if the process hasn't quit (we have to wait for some period to get stderr in the case of a failure).
I can't come up with a better name; anything with Simulator or Device in it will be confused with the existing IOSSimulator/Device classes (which represent the running devices).
This replaces the --prefer-shared-library flag, which falls back to
regular (non-shared-lib) compile if the NDK is not found, with the
--build-shared-library flag, which exits with an error message if the
NDK is not found.
This simplifies the set of allowed code paths through AOT compile,
resulting in better testability and easier-to-follow logic. It also
results in more predictable behaviour for continuous integration and
other scenarios.
Updates the message emitted when a kernel compile is skipped in the
build bundle action. Since we now use fingerprinting to enable
performance of script snapshots, AOT snapshots, and kernel compiles,
this helps a bit with debugging.
Also switches to trace level logging.
This change adds support for armv7, arm64, and universal iOS apps.
This change eliminates iOS target architecture hardcoding (previously
arm64 only) and uses the target architecture(s) specified in Xcode's
ARCHS setting ('Architectures' in Xcode Build Settings).
For universal binaries, set ARCHS to its default value, $(ARCHS_STANDARD).
Note that after changing the architecture in Xcode, developers should
run 'pod install' from the ios subdirectory of their project. A separate
change (that will land before this one) will add support for
automatically detecting project file and Podfile changes and re-running
pod install if necessary.
This change also adds an --ios-arch option to flutter build aot. In iOS
AOT builds (in profile and release mode), this dictates which
architectures are built into App.framework. This flag should generally
be unnecessary to set manually since flutter build aot is typically only
invoked internally by flutter itself.
This will allow external tools that wrap our test harness to share the
code that generates the test bootstrap.
This change exposed an issue whereby the LocalGoldenFileComparator
was being too strict in its URI handling, so this changes relaxes
that constraint as well (and adds associated tests).
If the developer changes their Xcode build settings and their project
has plugins, pod install is required, (e.g. to pick up changes to the
target architecture).
Similarly, manual edits to the Podfile should trigger a pod install.
Adds a Fingerprinter utility class that can be used to compute unique
fingerprints for a set of input paths and build options, compare to the
output of a previous run, and skip the build action if no inputs or
options have changed. The existing Fingerprint class still does all the
heavy lifting. Fingerprinter adds common operations such as
reading/writing/comparing fingerprints and parsing depfiles.
This migrates existing uses of Fingerprint over to Fingerprinter.
This also adds better fingerprinting to AOT snapshotting, which
previously failed to include several options in its fingerprint
(--preview-dart-2, --prefer-shared-library).
In order to avoid checking binaries into flutter/flutter,
this comparator can be used to retrieve golden files from
a sibling flutter/goldens repository.
https://github.com/flutter/flutter/issues/16859
Currently, `flutter update-packages --force-upgrade` will
crash if it encounters a pubspec.yaml file with no dependencies
(either regular or dev). The assumption that we'd never see
such pubspec files is no longer valid, as we have such a file
in one of our tests.
ProcessResult.stdout has static type dynamic so for
inference to infer proper type argument for the map
invocation we need to cast stdout to String explicitly.
Fixes#17163
Previously, we were incorrectly passing --vm_snapshot_data and
--isolate_snapshot_data options to gen_snapshot in assembly AOT builds.
These only make sense in AOT blob snapshot mode (alongside
--vm_snapshot_instructions and --isolate_snapshot_instructions).
There's very little code-sharing between the two, and what little there
is is concentrated in the GenSnapshotClass and the fingerprint
reading/writing utility methods.
This enables support for a `flutter_test_config.dart` configuration file,
which will be discovered and handed the responsibility of running the
test file (thus allowing it to run pre-test setup on a project level).
https://github.com/flutter/flutter/issues/16859
This de-duplicates assembly AOT configuration between Android and iOS,
and makes it easier to adjust parameters for 32-bit iOS (which, like
32-bit Android, requires --no-integer-division) in an upcoming patch.
iOS debug builds always run in interpreted mode whether on device or on
simulator. In both cases, we can skip snapshotting and link against an
empty App.framework. Previously, we did this for iOS simulator builds.
This does the same for device builds.
Previously, debug iOS builds used gen_snapshot to generate a core
snapshot, then used 'xxd' to generate C files containing the snapshot
data in buffers named kDartVmSnapshotData and kDartIsolateSnapshotData,
which are then compiled/linked into App.framework. This is unnecessary
since the VM compiled into Flutter.framework already contains this data.
Bugfix: Moves AOT snapshot input verification past where the last input
is added to the inputs list.
Cleanup:
* Extracts _isValidAotPlatform method.
* Moves non-platform-specific logic to the top.
* Moves variable declaration closer to first use, and inlines to a
narrower scope where possible.
This relands #17136, which was reverted in #17142 due to breakage in
on-device iOS debug builds.
Bugfix: Moves AOT snapshot input verification past where the last input
is added to the inputs list.
Cleanup:
* Extracts _isValidAotPlatform method.
* Moves non-platform-specific logic to the top.
* Moves variable declaration closer to first use, and inlines to a
narrower scope where possible.
* Add a `matchesGoldenFile()` async matcher that will match
a finder's widget's rasterized image against a golden file.
* Add support for pluggable image comparison backends
* Add a default backend that does simplistic PNG byte
comparison on locally stored golden files.
* Add support for `flutter test --update-goldens`, which will
treat the rasterized image bytes produced during the test
as the new golden bytes and update the golden file accordingly
Still TODO:
* Add support for the `flutter_test_config.dart` test config hook
* Utilize `flutter_test_config.dart` in `packages/flutter/test`
to install a backend that retrieves golden files from a dedicated
`flutter/goldens` repo
https://github.com/flutter/flutter/issues/16859
Moves the kernel compile step to the beginning of the AOT build in a
separate method. This is pre-factoring for iOS universal builds where
the kernel build happens once, but we then snapshot twice: once for
armv7 and once for arm64.
This also writes dependencies to build/kernel_compile.d rather than
build/aot/snapshot.d, since that is immediately overwritten by
gen_snapshot.
This is a simple refactoring with no functional changes. We now reuse
the existing _isBuildRequired() and _writeFingerprint() functions and
share them with script snapshotting rather than reimplementing their
logic.
Changes the signatures of both to support multiple output files (as
required for AOT snapshotting).
Fixes a bug introduced in 82f969ff05 where
the depfile used for AOT snapshotting, useful in particular for skipping
gen_snapshot when inputs/outputs haven't changed since the last build.
This is required for iOS debug builds, but unused otherwise. In theory,
Android debug builds could be run in this mode, but this is historically
untested and adds unnecessary complexity to the code. If ad-hoc testing
is required, it can be patched in when necessary.
This re-lands 8c4f0c0d21 with a fix to
xcode_backend.sh to eliminate the use of --interpreter.
This is required for iOS debug builds, but unused otherwise. In theory,
Android debug builds could be run in this mode, but this is historically
untested and adds unnecessary complexity to the code. If ad-hoc testing
is required, it can be patched in when necessary.
CompilerOutput values should never be mutated. Also includes minor
formatting change to the compile() method signature in line with the
style guide.
Minor housekeeping while I'm touching other bits of nearby code.
Wraps the compile function in a class injected via the global context,
which makes it easier to mock in unit tests -- specifically tests for
AOT snapshotting, which already require pretty significant amounts of
mock inputs.
New a11y traversal:
- sort direct sibling SemanticsNodes only
- use new sorting algorithm
- implement RTL
- test semantics in traversal order by default
- add AppBar traversal test
- breaking: remove nextNodeId/previousNodeId from the framework
- breaking: remove DebugSemanticsDumpOrder.geometricOrder
This avoids notifying the coverage collector that we
completed the test when in fact we timed out, which in
turn avoids a downstream exception caused by the fact
that coverage collector assumes it has a valid observatory
port by which it can gather coverage data.
Fixes https://github.com/flutter/flutter/issues/16839
This moves --vm_snapshot_data and --isolate_snapshot_data argument
hardcoding from GenSnapshot (a minimal wrapper around gen_snapshot
invocations) to Snapshotter.buildScriptSnapshot(). These arguments are
present in both AOT and script snapshots, but differ semantically: for
script snapshots they're inputs from the host engine artifacts
directory, for AOT snapshots they're outputs to the build directory.
* Support exposing the InspectorService over the Flutter
service extension protocol as well as the observatory protocol.
We will probably remove most of the observatory protocol support once a
couple versions of the Flutter IntelliJ plugin have shipped that use the
Flutter service extension protocol. The only reason to continue supporting
the observatory protocol is it will allow using the inspector when paused
at a breakpoint.
* Turn support for sync-async on in Flutter. With this feature
execution of async functions starts immediately instead of it
being delayed by one microtask tick.
* Address review comments.
* Adjust some stack trace expectations which are slightly different with
--sync-async turned on.
* Handle error count reported by frontend.
Extend compilation result from single string to a structure(string filename and integer error count).
* Use ?.
* Include engine roll with dart sdk roll.
* parse(onError) -> tryParse
* Make '?? throw' more readable and avoid issue with analyzer
* Fix test so it mocks compiler output including errors count
* A new UI is being designed to make the unused reload messaging clearer,
the UI will use the coverage data to highlight lines in the UI that were not executed during a reload.
In lieu of that we are turning off the unused reload messages experiment.
There were some issues in Dart2 also with this functionality
- source fingerprinting is not implemented in Dart2
- some additional synthetic functions are generated for invocation argument checking that are not appropriately filtered out.
Inlines the very small amount of work being done in _build() into
buildScriptSnapshot(). Eliminates a duplicate (and un-awaited) call to
_writeFingerprint.
* Sync .packages file as regular file.
Currently .packages file is treated specially as DevFSStringContent to accommodate package-file rewrite when it is sent to the device for dart1 compilation. In dart2 we need to treat .packages as regular file because from frontend perspective it's just a normal input file.
We are about to begin building gen_snapshot as a multi-arch binary,
which when run as x86_64 will generate arm64 AOT output, and when run as
i386 will generate armv7 AOT output.
Currently, gen_snapshot is an x86_64 binary, so this change is
effectively preventative in nature, and is a no-op with the current
snapshotter.
* Revert "Revert "Move mockito to 3.0.0-alpha. (#15949)" (#15979)"
This reverts commit e59651f925.
* More thenReturn to thenAnswer when mocking Futures
* Revert "More thenReturn to thenAnswer when mocking Futures"
This reverts commit 194d2cf417 as we are reverting engine roll.
* Revert "Revert "More thenReturn to thenAnswer when mocking Futures""
This reverts commit 52c9e96b30.
* Add dependency override
* Fix issue
* traverse dependencies and dev dependencies separately
* separate dev and reg deps
* update pubspecs
* add checksum string and update transitive dep string. Add additional verify command to check checksum
* cleanup comments
* add missing space, quote, run verify-onlu from bot script
* rerun tool with fixed string
* ensure correct working directory and bump versions
* flutter not flutterRoot
* roll engine version to 6280adbfb1f9f63cdc6179b9b78634add1e4f2e3
* Correctly handle previewDart : false setting in AnalysisServer and
AnalyzeOnce classes.
