This commit removes "is_inlined" flag from lsdump's VTableComponent,
because the clang library used to walk the ast doesn't report accurate
information about whether a member function is inlined or not.
Also update the testcases.
Bug: 80217511
Test: ./tests/test.py
Change-Id: I0e133b140a4f3debff8030de7d6712590cf98d76
- abi_dump.proto: Adds is_inlined and is_pure flag to VTableComponent.
is_inlined means a virtual function is inlined and probably not exported.
is_pure means a virtual function is pure and probably not exported.
Bug: 80217511
Test: ./tests/test.py
Change-Id: Ic1562ed47faba1f3d7eec6dffe94991296ba7176
- Convert method declarations to thunk symbols if thunk information
exists.
- Add a test case and reference dumps for virtual inheritance.
Bug: 80671888
Test: ./tests/test.py
Change-Id: I55ecbd7c8b0f5b0f534d81e6c6967f99ffd1d42e
Before this change, an elf symbol change not exported through public
headers would short circuit the check for api extension. As a result,
header-abi-diff would return a compatible return status (0). This isn't
correct, extensions should take precedence over elf only changes and an
extension status should be returned (non zero).
Test: tests/test.py
Change-Id: I6907d0ebaf332d019cdf2e264316f7d521e00d82
Earlier, we marked two records as different if their access specifiers
were not the same.
Bug: http://b/79493296
Test: tests/test.py
Change-Id: I117a408c7a85624a1e3d7a7477ae789eddd4bccc
The key being added to respective maps which facilitate diffing of
unreferenced user defined types was earlier the linker set key. The
unique id should be used instead. This is because linker set keys for C
and C++ types can be the same, but their unique id's cannot. Thus,
having linker set keys as a key leaves open the possibility of false
type aliasing, whereas the unique id does not.
Test: python3 tests/test.py
Test: python3 tests/test.py \
MyTest.test_libc_and_cpp_with_unused_struct_and_libc_and_cpp_with_unused_cstruct;
before this change, this test fails; after this change, the test
passes.
Change-Id: I91816167a48308a8c812d01e0a09e9bb70c60b50
The record kind (struct, class or union) was not being set while
converting from protobuf to IR.
Test: tests/test.py, all tests pass. Without the change, the default
kind of a C record would have shown up as class_kind.
Change-Id: I507581211a6d30c2c51414c6fab6413eece6b458
Previously, the tests mainly exercised header-abi-diff.
After this change, checked-in reference dumps are compared with
generated dumps. This allows us to:
- get header-abi-dumper and header-abi-linker involved in testing
better.
- Know about any incompatbile message format changes made
inadvertently.
Test: tests/test.py; all tests pass.
Change-Id: I7c7ba0811e6c08c6194a0a11616680279e7a0e1d
1) We do not involve the build system to create test reference dumps. Symbols
in tests, are faked through version scripts to avoid checking in prebuilt .so
test files.
2) Also add a script to create reference abi dumps.
3) Fix a comment in ast_processing.cpp
Test: Run tests/test.py, all tests run and pass on generic arm64, mips64
and x86_64 builds.
Change-Id: Id3a420c1ae093a05a79da4c29e85071f28d714d1
