Benchmarks on bullhead-userdebug show that there is a negligent
performance impact when using sparse entries on a 30% loaded
sparse type of 1000 resources.
Benchmark Time CPU Iterations
-----------------------------------------------------------------------------------
BM_SparseEntryGetResourceSparseLarge 255 ns 254 ns 2751408
BM_SparseEntryGetResourceNotSparseLarge 254 ns 254 ns 2756534
Bug: 27381711
Test: make libandroidfw_tests aapt2_tests
Change-Id: I051ea22f2f6b2bc3696e446adc9e2a34be18009f
libandroidfw needs to make use of StringPiece, so
move it to libandroidfw and update all code referencing
StringPiece in aapt2.
Test: make libandroidfw_tests libaapt2_tests
Change-Id: I68d7f0fc7c651b048d9d1f5e7971f10ef5349fa1
The multiwindow model and Resources-per-activity
model that came in N puts greater demands on AssetManagers.
They are created whenever window dimensions change, which
can be frequently. There is a need to be able to cheaply
create a new AssetManager for each Activity, which shares
a lot of underlying state.
In order to make the creation of AssetManagers cheap,
we need a new implementation of the native AssetManager
and ResTable to support immutable representations of
APKs. This new data structure/class is ApkAssets.
ApkAssets have the same functionality of an AssetManager, except
that they operate on a single APK, and they do not do any caching.
Once loaded, they are immutable.
ApkAssets will be exposed as a Java object, with its implementation in
native code. The existing Java StringBlock will be owned by ApkAssets,
which means that Strings can be shared across AssetManagers.
ApkAssets can be cached by the ResourcesManager. Creating an AssetManager
requires only a list of ApkAssets and a configuration.
AssetManager2 (named with the suffix '2' for now while transitioning
to the new implementation) caches bags that are accessed.
Since ApkAssets are expected to be kept around longer, they do more validation
of the resource table, which cause slower load times. Measured on an angler-userdebug,
loading the framework assets takes 11ms with ApkAssets, and 2ms with the old
AssetManager implementation.
The tradeoff is that there does not need to be any security checks once an ApkAssets
is loaded, and regular resource retrieval is faster. Measured on an angler-userdebug,
accessing resource (android:string/ok) with many locales takes 18us with AssetManager2,
and 19us with AssetManager (this is per resource, so these add up).
Test: make libandroidfw_tests
Change-Id: Id0e57ee828f17008891fe3741935a9be8830b01d
Tests would expect parts of the APK to be unzipped and
maintained. Instead, we now decompress the required files
from the test APKs on test setup. This simplifies
test maintenance substantially.
Test: make libandroidfw_tests && libandroidfw_tests --testdata=frameworks/base/libs/androidfw/tests/data
Change-Id: I3d2100af22df913e02401dedcf9842cdb32b2a3b
Instead of hardcoding or assuming a path for testdata,
allow the testdata path to be specified via the command
line.
Test: make libandroidfw_tests
Change-Id: Ideae880b21c157b70a11bb5a90a94556771aead6
- Adds unit tests for attribute resolution. These include
some test data resource tables and compiled XML files.
- Convert touched files to Google style guide.
Test: make libandroidfw_tests
Change-Id: Ib3a36061dc874de5f6a266b4e82c0a12ef435f23
A race could occur when printing the list of Asset allocations for
debugging purposes.
Each Asset object would insert themselves into a global linked list
on construction and remove themselves on destruction. Iterating the list
and the insertion/remove operations all acquire a global lock.
The race occurs after the Asset subclass destructor runs but before the Asset
base class destructor runs, which performs the actual removal from the list.
The vtable of the object being destroyed ends up pointing at the base Asset class'
vtable, and during the iteration of the global list, a pure virtual method is called
leading to an abort, since the wrong vtable is dereferenced.
This change moves the insertion/removal of the Asset object into the global list
to the concrete class, which adds some maintenance overhead but solves the problem.
Bug:31113965
Test: make libandroidfw_tests
Change-Id: I1a620897e5e04a8519ee247883bba0719b1fa6f3
- Added aapt command line flag --app-as-shared-lib to build app resources
that could be loaded as shared lib at runtime.
- Added new method AssetManager.addAssetPathAsSharedLibrary() to load an
app resource as shared library.
Bug 22487604
Change-Id: Ib9b33c35f9c2b7129f3ba205de03d4564623ea39
Was here because UBsan found integer overflow in the parsing for hex
numbers, since hex numbers here are actually unsigned but assigned to
a signed integer.
Also fixes a number of missing error conditions.
Change-Id: Iaea576daedfc6c75521cde02de3fe9dd0198a3b7
Shared libraries have their package ID assigned at run-time, so some
of the guarantees we used to have about sort order of attributes in
bags or XML elements no longer hold.
This CL adds back-tracking and can jump to the nearest attribute with the
same package ID and continue searching.
This means that attributes with the same package ID must be sorted by increasing
resource ID, as was the case before.
Attributes with the same package ID must be grouped together, but the groups can
be in any order. Ex: 0x02010001, 0x02010002, 0x01010000, 0x01010010, 0x7f010032
Bug:17666947
Change-Id: I9c198bbb6ca788849aac85b6323606ea5d9550d6
- char16_t is a distinct type, so stay consistent
with it throughout the code base.
- char16_t is defined as minimum size of 16 bits.
Since we mmap and cast data structures onto raw memory,
we need a precise definition (uint16_t), so we cast between
that (and static_assert that they are the same size).
Change-Id: I869c32637543bbcfb39d2643e7d9df10d33acd3c
The build system takes care of linking the correct STL and gtest
libraries for you, and specifying them manually confuses the build
system when using libc++.
Change-Id: I9f76ab26a63ace51293614cfb5ca002f37438e02
This is meant to be used with scaleable vector
drawables, and are chosen as the best match unless
there is a configuration that matches the density
requested exactly.
Bug:17007265
Change-Id: Ic3288d0236fe0bff20bb1599aba2582c25b0db32
In order to support APK split features, the resource
table needs to support loading multiple resource
tables with the same package but potentially new set
of type IDs.
This adds some complexity as the type ID space changes
from dense and ordered to potentially sparse.
A ByteBucketArray is used to store the type IDs in
a memory efficient way that allows for fast retrieval.
In addition, the IDMAP format has changed. We no longer
need random access to the type data, since we store the
types differently. However, random access to entries of
a given type is still required.
Change-Id: If6f5be680b405b368941d9c1f2b5d2ddca964160
- 3 letter codes are packed into the existing 32 bit locale
field in ResTable_config
- We introduce new fields for script / variant information.
Note that we define a "match" between two ResTable_config
structures to be purely on the basis of their language &
country (disregarding the script and the variant). However,
configs with scripts and variants are considered to be more
specific than those without.
(cherry picked from commit 378c6775a62d9c461cde51f06c1b14bb014c78fd)
Change-Id: I7dce82a3fe2412834252723f458826ae41535a78
- 3 letter codes are packed into the existing 32 bit locale
field in ResTable_config
- We introduce new fields for script / variant information.
Note that we define a "match" between two ResTable_config
structures to be purely on the basis of their language &
country (disregarding the script and the variant). However,
configs with scripts and variants are considered to be more
specific than those without.
Change-Id: I3c2e909ebbebdbbd0ab72a639d531b2b9fdbd9b9
This lets us share zip archive processing code with both
the runtime (Art, dalvik) and critical java code
(StrictJarFile).
This change also moves several utility methods to ZipUtils
and dedups code across several zip inflation methods.
One of the side effects of this change is that several
processing loops are now O(n) instead of O(n^2).
bug: 10193060
(cherry picked from commit afd31e08299008fdc5c2813f21b2573f29dc53df)
Change-Id: Iae67e62f1dc6dfc3f43e29bc38e3ffd1cb14d191
This lets us share zip archive processing code with both
the runtime (Art, dalvik) and critical java code
(StrictJarFile).
This change also moves several utility methods to ZipUtils
and dedups code across several zip inflation methods.
One of the side effects of this change is that several
processing loops are now O(n) instead of O(n^2).
bug: 10193060
Change-Id: I3c7188496837a47246c4f342e45485a70fef3169
No longer compile libandroidfw as a static library on the device
since it already exists as a shared library. Keeping the static
library would force us to provide a static library version of
libinput for the device as well which doesn't make sense.
Change-Id: I3517881b87b47dcc209d80dbd0ac6b5cf29a766f
Change the androidfw tests to the BUILD_NATIVE_TEST target so they end
up in the correct directory. Also remove the module tags and C include
paths. The include paths are automatically added when the library is
used.
Change-Id: Ia47f6c25130c5068b89d6dc067e5d9c714a6d08a
ZipUtils is needed by build/tools, move it from libandroidfw
(frameworks/base) to libutils (frameworks/native).
Change-Id: I2b4b7adcdf68eb25ee7cba5dd3b69eadf0523af3
