It was resulting in weird situations where the tool would dump an
error message and stack but not quit, or would fail hard but then just
hang.
Instead, specifically catch errors you expect. As an example of this,
there's one error we expect from the DartDependencySetBuilder, so we
catch that one, turn it into a dedicated exception class, then in the
caller catch that specific exception.
* Revert "Eliminate CocoaPods install step (#8694)"
This reverts commit f4a13bc72b.
If the developer is relying on CocoaPods and hasn't done a pod install, we will do it for them. This is needed for a smooth native plugin experience, similar to what Gradle is doing on the Android side.
There's no hard dependency on CocoaPods. We only run pod install if the project uses CocoaPods, so developers are still free to use alternatives if they prefer (and if they don't want to use native plugins).
Fixes#8685Fixes#8657Fixes#8526
* Require CocoaPods 1.0.0 or newer.
And make sure we don't get a crash if running `pod install` fails.
* Address review feedback
If the developer is relying on CocoaPods and hasn't done a pod install,
they'll get a build failure indicating the issue.
This also avoids a hard dependency on CocoaPods in the tool and allows
developers to customize their Xcode steps to use alternatives such as
Carthage if they prefer.
As of the latest Xcode versions, the latest published libimobiledevice
is out-of-date and build from HEAD is required.
This fixes two bugs:
1. Update initial install instructions to add --HEAD flag.
2. Update uninstall, reinstall instructions to include
--ignore-dependencies flag, since other brew formulae depend on
libimobiledevice.
Since iOS builds are CocoaPods enabled by default, we should make sure to run `pod install` to get pods wired up before building the app.
Also added a check to `flutter doctor` to verify CocoaPods is installed.
I'm passing FLUTTER_FRAMEWORK_DIR to the `pod install` command, so we can have the app's Podfile link in Flutter.framework as a pod instead of having to copy it over in xcode_backend.sh.
This implements the `DartDependencySetBuilder` completely in Dart instead of calling out to `sky_snapshot` (Linux/Mac) or `gen_snapshot` (Windows) and allows us to use the same code path on all supported host platforms.
It also slightly reduces hot reload times on Linux from ~750ms to ~690ms for the unchanged flutter_gallery app and significantly reduces hot reload times on Windows from almost 1.5s to just slightly slower than on Linux.
This change will also allow us to retire `sky_snapshot` completely in the future.
* add crash reporting without enabling it
* do not drop futures on the floor
* return exitCode from executable run
* debug travis
* remove unnecessary todo
* rename local fs getter
* Enable Hot Reload on Windows (backed by gen_snapshot)
\o/
Two caveats:
* Hot Reload on Windows is slower than on other platforms because gen_snapshot is slower then sky_snapshot
* We currently cannot hot reload projects with spaces in the path
* enable tests
The gradle build scripts can be configured to output different
application IDs for different build types, so we need to examine the
built .apk to figure out the name of the package and activity.
Re-landing this change, updated to only get information from the .apk
if it exists.
Since the tools create an AndroidApk instance early, even before we've
actually built an .apk, we have to create a new instance after building,
so we can start the right app/activity.
Fixes#8327.
* [devFS] Use URI to represent paths on device
Previosuly, regular file paths in the format of the host platform were used to represent paths on device. That works when host and device share the same (POSIX) file path format. With a Windows host, this breaks. URIs are the solution as they are platform independent and the VM service on the device already interpreted the file paths as URIs anyways.
* review comments
* switch to file paths
* fix tests on Windows
* review comments
* Revert "Revert "Simplify path handling logic in dependency checker and devFS (#8414)" (#8467)"
This reverts commit 96ba7f76d2.
* Intentionally use a self-package URI in flutter_gallery
* tests to catch problems with self-package imports
This adds support for a `--bug-report` flag, which is a recording
that:
- includes the arguments that were passed to the command runner
- is zipped up for easy attachment in Guthub issues
* Simplify path handling logic in dependency checker and devFS
Simplification will make it easier to port this to Windows.
* Roll Engine to 0a7b177c330367904597a6129b3eb653d29dfca0
The gradle build scripts can be configured to output different
application IDs for different build types, so we need to examine the
built .apk to figure out the name of the package and activity.
Fixes#8327
It's not just $HOME/.AndroidStudio2.2, it might also be
.AndroidStudioPreview2.3, or .AndroidStudioFooBar1.7, or whatever.
Made the Version parser less throw-happy, and relaxed the directory name
checks to allow for the above.
Fixes#8353.
If a target file is specified on the flutter tools command line, pass it
through to Gradle.
It is still possible to statically specify a target file in the flutter
section of build.gradle, but it is now possible to specify it on the
command line as well. The command line option takes precedence.
Fixes#8175.
Only implemented for Android devices for now. Compare the installed SHA1
to the latest build. If they match, there's no reason to reinstall the
build.
Fixes#8295
* Fix bug parsing Gradle version.
Version from pub_semver requires versions of the format X.Y.Z. Gradle
doesn't follow semantic versioning, though, so version parsing would
fail on versions like '3.2'. Fixed by writing a custom Version class.
Also removed a check for apksigner when building Gradle-based projects.
Fixes#8298
Do not attempt to deploy/debug wifi connected iOS devices. ios-deploy is
able to install over wifi, but we've had several bugs reporting failure
to run/debug once installation has completed when the device is also
connected via USB. Note that we don't currently support deploy/debug
over wifi since libimobiledevice (which is also required) requires a USB
connection.
* Remove SHA1 check from AndroidDevice.isAppInstalled()
The docs for isAppInstalled say 'check if a version of the given app is
already installed', however the current code returns true only if it's
the latest build that's installed.
This made sense in the past, when the use pattern was:
if (!isAppInstalled(...)) installApp(...);
but now the usage is:
if (isAppInstalled(...)) uninstallApp(...);
installApp(...);
This has the probably unintended consequence that if you run `flutter
install` or `flutter run` two times in a row with no source changes, the
second invocation will uninstall the app, but the first invocation might
not.
Removing the SHA1 check makes us always uninstall the app if it's
installed.
Fixes#8172
* Teach flutter tools to find gradle
Flutter tools will now use Gradle from Android Studio, which is now found automatically.
flutter doctor will verify that Android Studio has been installed, and that the included Gradle is at least version 2.14.1.
It is still possible to manually configure the path to Android Studio (flutter config --android-studio-dir=XXX) or Gradle (flutter config --gradle-dir=XXX), but this should only be necessary if they're installed somewhere non-standard.
Only tested on Linux and macOS for now.
Fixes#8131
Artifacts are now located in a central place.
This will enable us to downlaod artifacts when we need them (instead of
downloading them all upfront).
This also makes replacing sky_snapshot with gen_snapshot easier.
If a developer has installed the Xcode command-line tools, then later
installs Xcode, they may need to run xcode-select to update the path of
the command-line tools to that of the new installation.
On flutter run, we update ios/Flutter/Generated.xcconfig with various
Flutter-specific settings required by xcode_backend.sh during a build
from Xcode. These settings need to be present at the time the project is
loaded since Xcode doesn't pick up live updates to these files.
Without these settings, Xcode fails to locate xcode_backend.sh itself,
causing the build to fail until the Xcode project has been closed and
re-opened. This also prevents Xcode's project updater from 'helpfully'
suggesting to clean up and delete the Generated.xcconfig file.
* Make new project template gradle-based for Android.
With this change, the 'new project' template uses the same gradle-based build for Android as the hello_services example. This has some implications on build performance, since we're now building a complete Android app instead of just combining a pre-compiled .dex with the Flutter assets.
The very first build is a little over 2x slower, since it needs to download gradle and build the build scripts before getting to the actual code. Subsequent builds with changes to the code are comparable to the old builds. Null builds are faster. Enabling the gradle daemon speeds up subsequent builds by around 5s.
* Move Flutter Gradle plugin to Flutter root.
* Fix potential NPE in iOS doctor check
In case Xcode is not installed, the xcode-select path may be null.
* fixup! Fix potential NPE in iOS doctor check
If the developer has only installed the Xcode command-line tools,
xcode-select and some other tools may be present, but xcodebuild will be
missing. In this case, output a slightly improved message indicating
that the installation is incomplete rather than missing.
* Small Flutter strong mode cleanup fixes.
These are cases where strong mode down cast composite errors
generally indicated cases that would performance or correctness
issues if Flutter code was run in a strong mode VM.
* Fix Command API so that it is always in terms of Map<String,String>.
* Fix typedef
If ensureDirectoryExists fails -- e.g. because a file file of the same
name as the directory to be created exists, ensure that we exit cleanly
with a useful error message.
If the tool is started immediately after a previous run of the tool, then
the most recent log timestamp may come from the previous run's logs.
The new instance of the tool should not print those logs.
This was happening during runs of the microbenchmark script.
* Add doctor check for Python 'six' module
Required as part of Xcode lldb module. In all likelihood, if we
encounter this situation, the developer is using a custom Python install
(e.g., via MacPorts or Homebrew).
* Return null from VM.mainView if no view exists
* Retry in connectToServiceProtocol if a view is not yet available
* Do not explicitly call exit from the benchmarks - it will not cleanly shut down the engine
* Update gradle example to support x86 in debug mode.
Changed the Flutter Gradle plugin a bit to better fit in with the
Android build.
Fixes#6136Fixes#6864Fixes#7539
This remove a very brittle aspect of flutter drive, whereby it would
assume a known port instead of explicitly finding out what it was.
Fixes#7692 and hopefully fixes the devicelab tests.
Move the drive/run common stuff to RunCommandBase, rather than the
superclass which applies to everything.
Implement --no-resident (fixes
https://github.com/flutter/flutter/issues/5448). This is particularly
useful when running flutter from scripts.
Rename RunAndStayResident to ColdRunner.
* Add a check in case the flutter directory in .packages no longer exists. Clean up and prompt user
* Update documentation to use flutter packages get for end-users instead of flutter update-packages.
* Merge missing sdk error with the multiple sdk error. They're really the same thing.
* Use flutterPath in both checks.
* Change file_system’s copy folder to copy director which takes into account the file system
* Test support files
* Add test and split into 2 messages again.
* Move tests to run in memory file system's copy. Tested with dev/bots/test.sh
The zip files downloaded from cloud storage are not clearly separated
into engine artifacts and tools anyways.
Also, because of this unclear separation, we were downloading
`linux-x64/artifacts.zip` twice before.
* Change file_system’s copy folder to copy director which takes into account the file system
* Address review comments.
Test with 2 different instances of file systems.
Refactor DevFS so that it's easier to add new types of content such as kernel code
* add tests for DevFS package scanning
* add tests for DevFS over VMService protocol
* which covers _DevFSHttpWriter and ServiceProtocolDevFSOperations
* replace AssetBundleEntry and DevFSEntry with DevFSContent
* refactor to cleanup common code and replace some fields with locals
* rework .package file generation refactor away DevFSOperations.writeSource
* only update .package file if it has changed
* only write/delete/evict assets that have been changed/removed
Also, defer to test package for throttling (this will require a test
package update as well).
Also, add a lot more instrumentation to --verbose mode for tests, and
fix minor trivial things here and there, and add error handling in
more places.
Also, refactor how coverage works to be simpler and not use statics.
When a machine is heavily loaded, it can severely delay the time it
takes for the OS to start a process once it's asked to do so. Our
5 second timeout that we were giving the test process to connect
to the test harness seemed like plenty of time, were it not for the
fact that the test process itself was not being started in time when
CPU throttling was in effect.
This change updates the test timeout to begin counting only once the
test process has been started. We keep the original timeout in play
in the event that the test process *never* starts up for some reason,
but we up that timeout value to 5 minutes.
Stop building (unused) unlinked summaries for packages.
Improves update speed considerably (for `n` packages it saves us `n` needless calls to `pub get`).
Previously, it was possible for the test harness to bail
and the test runner to complete before the platform plugin
triggered the collection of coverage data. This fixes the
race condition such that the pending coverage collection
task is recorded immediately after starting the process.
As of Dart SDK 1.22.0-dev.5.0, `Process.exitCode` is no longer
mutable (that SDK version picks up e5a16b1ca5).
This change allows the tools code to pass analysis in sdk versions both
before and after that change, to allow for analysis against both the host and
target sdks.
We now produce a more reasonable error message when we're missing the
flutter_test dependency in a test. Also, remove the flutter_tools stack traces
when the engine dies.
Fixes#6187
Each stack trace will yield many async task stacks for every write request
that is in flight. If the device side is unresponsive and all writes are
failing, then this can generate an overwhelming amount of logs.
Fixes https://github.com/flutter/flutter/issues/7351
When a test runs to completion, the test harness closes the stream side of the
StreamChannel, causing the sink side to be closed as well. So by the time we
receive a test result of completed/harnessBailed, the controller sink has been
closed.
This removes direct file access from within flutter_tools
in favor of using `package:file` via a `FileSystem` that's
accessed via the `ApplicationContext`.
This lays the groundwork for us to be able to easily swap
out the underlying file system when running Flutter tools,
which will be used to provide a record/replay file system,
analogous to what we have for process invocations.
