The flag `disabled_rw_2` is used to verify that aconfig can
auto-generate code for flags in different namespaces. Rename the flag to
`disabled_rw_in_other_namespace` to highlight that it belongs to a
different namespace than the other flags.
This CL is a semantic change only.
Bug: None
Test: atest aconfig.test
Change-Id: Ib82fdce0ca3f7cd56b9b3f80e44fe424a03a2ae4
This reverts commit efda207c09.
Reason for revert: this unrevert contains the fix for b/311187402
Test: cargo test
Change-Id: I210aebd30edd864a7c141ede336c12aebf4f1fcd
Improve performance slightly: remove unnecessary clone operations, or
use references where a new object is not needed.
Bug: 283910447
Test: atest aconfig.test
Change-Id: I75205ffa1723dd2654039baac882c225d2653c86
Add a new field in the declaration to indicate whether
the permission can be overridden.
When the field “is_fixed_read_only” is set to true,
the flag permission will be set as fixed “READ_ONLY”,
and the permission should not be changed by Gantry.
Bug: 292521627
Test: atest aconfig.test
Change-Id: Ic9bcd7823bccb8b947cf05568c7ced3763490a23
Update sys prop name from persist.device_config.<namespace>.<flag> to persist.aconfig_flags.aconfig_flags.<namespace>.<flag>
Bug: b/295379636
Change-Id: I78f00daa574dc016902dce4b9b4ec961f973693c
1, Moved "#include <string>" from exported header to test flag provider
header file.
2, For production target and read only flags, the generated c
api should just return default value instead of calling into c++ api
3, Remove using namespace server_configurable_flags from header, instead
of just having the namespace to be spelled out in each api call. Having
using namespace xxx in header is not a c++ best practice.
4, Replace #ifdef #def #endif with #pragma once
Bug: b/279483801
Test: atest aconfig.test
Change-Id: I3e55a7b14301f3de419795467f33e2dc889d371e
Two major changes to c/c++ codegen
(1) explicit setter for each flag instead of a generic flag setter
void override_flag(std::string name, bool val) is replaced with
void <flag name>(bool val) for each flag name
This has several advantages:
(a) generated code is more c++ idomatic
(b) no longer need to create flag name string constants
(c) any typo in the code is caught early in the build time
(2) remove flag setter and flag override reset methods/functions when
generating code targets for production. If developers want to update
their main function to take command line arg for flag overrides, they
can use compile time macros to decide if the flag override code should
be included.
Bug: b/279483801
Test: atest aconfig.test
Change-Id: I6141f7f979b32fe0426154d578edeb997ae5ff7c
c codegen can be done at the same time with cpp codegen, the idea is to
create a c compatible header that defines the flag apis, including flag
override apis for test. then in the corresponding cpp file, the
implementation simply calls into cpp api.
c header supports static method interface, and unit test override, but
it does not support injection pattern compared to cpp header
Bug: b/279483801
Test: atest aconfig.test
Change-Id: Ie62b76d6524e443de5d3c2f9000f7f66623ab571
Move server_configurable_flags header include away from exported header,
instead put it in the flag provider headers. Otherwise, it can cause
compilation error if the source code who wants to use the generated c flag lib has
not added dependency on server_configurable_flags yet.
Bug: b/279483801
Test: atest aconfig.test
Change-Id: Ib75877ee88f995caf72b3fd2554c3da195032c14
cpp codegen iteration 2, based on discussions with three internal teams
that use c++. Refer to the design doc "aconfig c++ codegen" for detailed
design. At a high level, we generate two sets of code artifacts with the
same signatured api: one for production that without any local flag
override capability, one for unit test that allows local flag overrides.
It supports static methods style interface as well as injection pattern.
Refer to the test points in the codegen_cpp.rs for examples of generated
code.
for production target codegen: aconfig create-cpp-lib --cache <cache> --out <out dir>
for test target codegen: aconfig create-cpp-lib --cache <cache> --out
<out dir> --mode test
Bug: b/279483801
Test: atest aconfig.test
Change-Id: I92fefb9623d5435525339a74f57bbd36d0afef08
Remove the hand-crafted wrappers around the structures auto-generated
from protos/aconfig.proto, and use the auto-generated structs directly
intead. This gets rid of a lot of manual repetition, and its inherent
risk.
Also unify how individual fields read from text proto are verified (e.g.
is the flag.name field a valid identifier).
Also change the intermediate cache format from JSON to binary protobuf.
The concept of a 'cache' as an intermediate internal format to represent
parsed input stays. The command line interface still refers to caches.
At the moment a cache file is identical to a parsed_file protbuf, and
the code exploits this internally.
A couple of points regarding the auto-generated structs:
- Vectors are named in the singular (e.g. parsed_flags.parsed_flag is
a Vec<ProtoParsedFlag>) because this improves ergonomics for all
devs working with aconfig input files
- The auto-generated structs have fields that are of type Option<T>
and convenience methods (named the same as the fields) to access T
Test: atest aconfig.test aconfig.test.java
Bug: 283910447
Change-Id: I512820cc4bc6c543dea9f6a4356f863120a10be3
Implement a helper function to make it easier for unit tests to diff
(and find the first difference) generated code and expected code.
Bug: 283910447
Test: atest aconfig.test
Change-Id: I460e8fbf05e8f33e8a62ecef67b2d9d77051e876
Add a new field to the proto messages flag_declaration and parsed_flag.
The new field will be used verbatim as a parameter when calling
DeviceConfig.getBoolean to read the value of a READ_WRITE flag. See the
DeviceConfig API for more info.
Note: not to be confused with the old namespace field, which has been
renamed to package.
Bug: 285211724
Test: atest aconfig.test
Change-Id: I2181be7b5e98fc334e5277fb5f7e386f1fe0b550
Allow package fields to include dots.
Update the generated code based on the package name: if the package name
is com.android.example:
- java: package com.android.example; ...
- C++: namespace com::android::example { ... }
- Rust: mod com { mod android { mod example { ... } } }
Also, update examples to use dots in the package fields.
Also, remove unnecessary #include from the auto-generated C++ code: the
header should not include itself.
Bug: 285000854
Test: atest aconfig.test
Change-Id: I8a5352e25c64c34dee0725202a1b7c9957819de8
What used to be referred to as a namespace is now called a package.
This CL is a semantic change only.
Bug: 285000854
Test: m nothing
Test: atest aconfig.test
Change-Id: If3fca67c415af75b44f316e16666b97089407069
Introduce a builder pattern for constructing a cache from flag
declarations and flag values. Teach the builder to sort the flags by
name as the last step. This will ensure consistent dump output
regardless of the order flags are specified in the input files.
Bug: 279485059
Test: atest aconfig.test
Change-Id: Icdd62f51fa3761a469663f17581a83d9909e9ffe
The general idea to reuse java codegen's very neat tiny template idea.
For generated cpp code, it is in the form of a collection of classes
inside a namespace. The reason we choose a collection of classes rather
than a collection of static functions is because gmock test technology
only supports mocking virtual method.
Bug: b/279483801
Test: atest aconfig.test
Change-Id: I9ba00667437ff7c3e147ff2828171fc95528bebf
