Note: If there‘s anything you want to see done in the NDK, file a bug! Nothing here is set in stone, and if there’s something that we haven‘t thought of that would be of more use, we’d be happy to adjust our plans for that.
Disclaimer: Everything here is subject to change. The further the plans are in the future, the less stable they will be. Things in the upcoming release are fairly certain, and the second release is quite likely. Beyond that, anything written here is what we would like to accomplish in that release assuming things have gone according to plan until then.
Note: For release timing, see our release schedule on our wiki.
Estimated release: Q4 2018
Now that the NDK is down to a single compiler and STL, if we just taught the Clang driver to emit -D__ANDROID_API__=foo and to link libc.so.18 instead of libc.so, standalone toolchains would be obsolete because the compiler would already be a standalone toolchain. The NDK toolchain would Just Work regardless of build system, and the logic contained in each build system could be greatly reduced.
Related to this (but maybe occurring in a later release), we'll want to switch from libgcc to libcompiler-rt and our own unwinder.
See the corresponding bug make all toolchains standalone toolchains for detailed discussion of the implementation and sub-tasks.
Estimated release: Q1 2019
To be decided...
NDK r17 added names for all function arguments, but tools such as vim and Visual Studio Code need a compile_commands.json file.
We should probably add basic doc comments to the bionic headers:
char* point to? Who owns it? Are errors -1 (as for most functions) or <errno.h> values (for pthread_mutex_lock)?Should these be in the NDK API reference too? If so, how will we keep them from swamping the “real” NDK API?
vim is ready, Android Studio is almost ready bar one bug (https://issuetracker.google.com/110556794), and Visual Studio Code has nothing but feature requests.
The samples are low-quality and don't necessarily cover interesting/difficult topics.
The NDK has long included gtest and clang supports various sanitiziers, but there are things we can do to improve the state of testing/code quality:
There are many other commonly-used libraries (such as Curl and BoringSSL) that are currently difficult to build/package, let alone keep updated. We should investigate using something like cdep to simplify this.
NDK r18 made lld available, but we should make it the default (as it already is in the platform), and long-term aim to ship lld as our only linker. https://github.com/android-ndk/ndk/issues/683
We should make lldb available in the NDK. It's currently shipped as part of Studio.
Are modules useful and is the clang implementation complete enough? How do we test? Is this only useful for libc/libm/libdl or for the NDK API too?
NDK APIs are C-only for ABI stability reasons. We should offer header-only C++ wrappers for NDK APIs, even if only to offer the benefits of RAII.
Complaints about basic JNI handling are common. We should make libnativehelper or something similar available to developers.
For serious i18n, icu4c is too big too bundle, and non-trivial to use the platform. We have a C API wrapper prototype, but we need to make it easily available for NDK users.
We still need to update libc++ twice: once for the platform, and once for the NDK. We also still have two separate test runners.
CMake added their own NDK support about the same time we added our toolchain file. The two often conflict with each other, and a toolchain file is a messy way to implement this support. However, fully switching to the integrated support puts NDK policy deicisions (default options, NDK layout, etc) fully into the hands of CMake, which makes them impossible to update without the user also updating their CMake version.
We should send patches to the CMake implementation that will load as much information about the NDK as possible from tables we provide in the NDK.
iOS developers are used to using weak symbols to refer to function that may be present in their equivalent of targetSdkVersion but not in their minSdkVersion. They use a run-time null check to decide whether the new function is available or not. Apparently clang also has some support for emitting a warning if you dereference one of these symbols without a corresponding null check.
This seems like a more convenient option than is currently available on Android, especially since no currently shipping version of Android includes a function to check which version of Android you're running on.
We might not want to make this the default (because it's such a break with historical practice, and might be surprising), but we should offer this as an option.
Full history is available, but this section summarizes major changes in recent releases.
Removed GCC and gnustl/stlport. Added lld.
Defaulted to libc++.
Removed ARMv5 (armeabi), MIPS, and MIPS64.
Fixed libandroid_support, libc++ now the recommended STL (but still not the default).
Removed non-unified headers.
Defaulted to unified headers (opt-out).
Removed support for API levels lower than 14 (Android 4.0).
Added unified headers (opt-in).
Added simpleperf.
Removed armeabi-v7a-hard.
Removed support for API levels lower than 9 (Android 2.3).