The checked-in libclang_rt.* prebuilts do not contain versioned symbols,
but the autogenerated map.txt file for these libraries contain versions
e.g.
LIBCLANG_RT_ASAN {
global:
...
}
```
This causes a discrepancy between the symbols in the stubs and impl
variants of these libraries. After https://r.android.com/3236596 is
submitted and a new set of libclang_rt.* prebuilts are generated, this
discrepancy will go away.
For now, skip `-Wl,version-script` when generating the stub
variants of these libraries to prevent link errors at build time.
Test: lunch aosp_panther_hwasan-trunk_staging-userdebug && m crash_dump
Bug: 3222365
Change-Id: Ic1e06cb58ec33b761136ac325f191da7ac900151
This reverts commit 608779b224.
Reason for revert: Droidmonitor created revert due to build breakages in b/362287839.
Change-Id: I6ff96539ae714698692794209d8e76e6062378b0
IsNdk and NotInPlatform can return incorrect values in minimal branches
that do not contain the source bionic libraries. (e.g. in
aosp-master-art)
Bug: 275273834
Test: presbumits (which will build aosp-master-art)
Change-Id: I5e9216a40c0fd2d077bb6c53ba6628e311d99737
So that you can use selects with them.
Bug: 342006386
Bug: 358377461
Test: m nothing --no-skip-soong-tests
Change-Id: I5a8350f670d51b7da411ad5c3cdbf5f2d6cdd63b
The `stubs` property in `cc_prebuilt_library` currently does not support
generating stubs from an API symbol file. Reverse dependencies are
always linked against the .so file listed in `srcs`. This CL adds
support to generate the shared library from the .map.txt symbol file of
the prebuilt library.
Implementation details:
1. Add a compiler. This will convert the .map.txt file to a stub .c file.
The build rules for this are common to the compiler of the source cc_*
moudule types. This has been refactored into a new function named
`compileModuleLibApiStubs`
2. Update the linker to link the aforementioned stub.c file into a .so
file. Since this happens only for stub variants, it is conditionally
called only when p.buildStubs is true.
3. Call `addStubDependencyProviders`. This will populate the impl
variant of the library with Stub information. cc#ChooseStubOrImpl
will use this information to resolve rdeps to the stub .so file if the
dependency crosses an api domain boundary.
(2) requires some special casing for now, since the module sdk at HEAD
does not contain symbol files, so the stub.c files will be empty. Once
the symbol files are added to module sdk, the special casing will be
removed.
Test: m nothing --no-skip-soong-tests
Bug: 275273834
Change-Id: I57f5b0a97fac4dfc8c24e07b06a330015add977d
Adding stub generation support for C prebuilt libraries creates
duplicate dist rules for api coverage files
```
overriding commands for target `<out_dir>/../cc_apis/libc.xml', previously defined at
```
This CL hides the incorrect api coverage file using `IsModulePreferred`.
Bug: 275273834
Test: EMMA_INSTRUMENT=true EMMA_INSTRUMENT_FRAMEWORK=true CLANG_COVERAGE=true NATIVE_COVERAGE_PATHS="*" m droid dist
(no duplicate dist rules)
Change-Id: Id1caade7e80119ed1dfbbef638feb22adf116d63
This is to limit the direct accesses to the internal fields of a
module in order to better support incremental caching. To access
the install files data from singleton context or other modules'
context use providers thru the provided wrapper; to access it
from the same module inside GenerateBuildActions use ctx which is
short-lived only inside this method.
Bug: 358425833
Test: CI
Change-Id: I337b07a2ef95fb2a898ac2f9277160a3f76a603c
We now have a modern libc++, and no longer needs to explicitly turn off
deprecated-dynamic-exception-spec warning for it.
Test: presubmit
Change-Id: Ie282b32ed9ac2b0246fd8245f4c5d29526cbda76
This is a useful warning that can point out lower performance code.
https://r.android.com/q/hashtag:pessimizing-move-cleanup cleaned up many
cases. Let's bump this up to a non-error warning while we fix up the
remaining cases.
The warning remains off for all external code.
Test: presubmit
Bug: 154270751
Change-Id: If6b69cbaa1384e81336b6842607a587bf6c7f38b
Add a helper function that creates a test fixture preparer that
sets a build flag, and use it everywhere that was setting build flags
manually.
Test: all soong tests
Flag: EXEMPT refactor
Change-Id: I68d50d68787a30d091f0827e8caa51f5c5a762ef
This is being done with a path-based filter rather than as a property
of ndk_library because there would be no way to prevent people from
disabling ABI tracking without API council review if it were a
property, since there's no per-module OWNERS.
Bug: http://b/358653811
Test: m ndk
Change-Id: If6af638accc917294eee18cb08551c5df25462ad
This avoids allocating an intermediate buffer in memory *and* avoids an
unchecked error on `(*File).Write` which can mask partial writes.
Check error on `(*File).Close` while I'm here.
Change-Id: I9f7c70f1eb267155c080e44c50a99bdca62dea05
Revert submission 3207854-ensure-c-ndk-headers
Reason for revert: Droidmonitor created revert due to b/358445530.
Reverted changes: /q/submissionid:3207854-ensure-c-ndk-headers
Change-Id: I649941a4d063d76a1c6a8e58fb885da2c44331c0
https://github.com/android/ndk/issues/1920 recently reported a handful
of NDK headers where C-incompatible syntax had slipped in which wasn't
caught by review. This is also something that we often manually catch
in API review, but the manual inspection is, as can be seen from that
bug, error prone.
This check is pretty stupid. I've tried on other occasions to do a
more thorough check that would build each header using the same flags
as cc_library or similar would, but was never able to get that
working. This isn't as thorough, but that's maybe not such a bad
thing, since it saves on build time to only check on variant. In
practice the rudimentary approach I've taken here would have caught
every instance reported in the bug above, and probably all the other
cases that we'd previously caught in API review as well, so it's
definitely a step in the right direction if not good enough forever.
Bug: http://b/113359184
Test: m ndk
Change-Id: Id4e8cc511413acc61c4f625f25c3004d7439263c
This works around an incompatibility between coverage and the
-fdebug-info-for-profiling flag.
also moved -fdebug-info-for-profiling flag so that it is only applied to
the libraries that enabled AFDO.
Test: gzip -cd out/verbose.log.gz | grep debug-info-for-profiling
Bug: 345593672
Change-Id: I68493511da1e61091209d0ed1b2c86c7ba0e21f4
It doesn't need to be a Makefile variable exported from Soong as it's
hard-coded.
Bug: 353429422
Test: boot shiba with SANITIZE_TARGET=hwaddress
Change-Id: I4a98eaaf1002aa7aba5d5131ff251bdcbdd2e0ef
Update compiler to r536225 and the error: argument '-fno-relaxed-template-template-args' is deprecated arises.
Bug: b/341084395
Test: presubmit
Change-Id: If714f96e4537b5f8d695cd7c0d1cd0a8ddcbb96e
To support skipping ninja file writing during incremental build, we
add all the global rules/variable/pools from static sources because
when we restore from cache we don't have the build defs available to
build the globals. This means all the global rules and variables are
added blindly, so we need to exclude these darwin specific ones since
they will fail due to they depends on the tool that only exists on MAC.
Bug: 335718784
Test: local tests and CI
Change-Id: Ife835545f0f76ffaec77e48bdd89ab7384a1b995
