* Run frontend server from temp directory to force absolute URIs in Kernel file.
* Create special directory to cwd into to avoid conflicts with app sources in temp folder
* Make variable final
As we anyway always recommend to set the target_os to android.
Also if you happen to not set the target_os to android it messes up the licenses script output.
* Move texture registry ownership to platform view
This enables the texture registry to survive activity pause on Android.
* Remove debug info
* Formatted
* Set texture registry on initial rasterizer
* Remove unneccessary std::move
If Progress is instantiated with a max of 0, throw immediately to avoid
a divide-by-zero later in toString(). This typically happens if the tool
recurses over an empty top-level component, which can happen when a
component is moved around in the repo and the developer hasn't cleaned
up old empty directories from their git client.
* Enable flutter engine to also work with .*so files on android
We would like to be able to use native tools (e.g. simpleperf, gdb) with
precompiled flutter apps. The native tools work much better with *.so
files instead of the custom formats the Dart VM uses by default.
This CL adds support for being able to load the flutter app from an *.so
file on Android.
* Add sanity check to ensure we have either shared library or instruction snapshot (but not both)
The new ui.Codec API is a superset of the implementation in
image_decoding.cc. While we are not using ui.decodeImageFromList in the
framework anymore, we keep the dart function for backward compatability, and
just make it a thin wrapper around the codec API.
When checking whether render buffer size has changed, apply the device
scale factor used to convert points to pixels to ensure we're not
needlessly re-allocating render buffer storage.
Adds trace + debug log to make this easier to detect in future.
Originally introduced in 72c28f3a01bd9e6bed256edc4629705d96a7cead, but
rolled back in d539abd7c5fe5291bab81b32e6e5697c1b6681b6 to test whether
it was responsible for regressions to
flutter_gallery_ios__transition_perf benchmarks:
* average_frame_build_time_millis
* missed_frame_build_budget_count
Re-landed in 1a7ca91e3938abd5720b3553dfb2045a6b805a3b, and reverted in
79b9bdbbf4d3193620be35e3ea3e1a162b553627 after it appeared to regress
the above two benchmarks. After recent changes to microtask execution
order which fixed https://github.com/flutter/flutter/issues/9998,
re-introducing this fix results in benchmark improvements to the above
two benchmarks in local testing.
The UIDartState is now always owned by the isolate and always freed in
the isolate cleanup callback.
In the isolate shutdown callback, if the isolate being shut down is the
main isolate, the RuntimeController is informed which in turn notifies
the RuntimeHolder and thus the ApplicationControllerImpl. The
ApplicationControllerImpl tears down the whole Flutter application.
This fixes Fuchsia bug: MI4-328
Fixesflutter/flutter#13018
When linear blending was disabled, we started rendering directly to FBO0 again. We can't attach stencil there, and the profile graph code triggers a path that (by default) uses it. This option forces us to use alternate rendering methods.
Note that the graph rendering code is constructing a fairly complex path. It would probably be much faster to render as a simpler series of drawRect calls for each box (which would get batched inside Skia).
This retains gamut correction (adjusting colors for screens with different capabilities), but does all blending and interpolation with sRGB-encoded values. That matches the behavior expected by most users, as well as the behavior of nearly all other systems. It also greatly simplifies the EGL code.
A future Skia change will make this behavior more of a first-class citizen, so some of these implementation details will change again, but the behavior will not. The bulk of this change (elimination of complication from the GL surface code) is permanent - it's just the SkColorSpaceXformCanvas that will be replaced.
