Newer versions of the macOS SDK, have introduced code like:
```cpp
if defined(__has_feature) && __has_feature(modules)
define USE_CLANG_TYPES 1
else
define USE_CLANG_TYPES 0
endif
if USE_CLANG_TYPES
include <sys/_types/_ptrdiff_t.h>
include <sys/_types/_size_t.h>
include <sys/_types/_va_list.h>
include <sys/_types/_wchar_t.h>
endif
```
which is currently causing compile failures due to undeclared types,
which manifest when C++ modules are enabled. Note that the usage of
"modules" in LLVM is can be a bit ambiguous, see:
https://github.com/llvm/llvm-project/issues/55891.
For now, explcitly disable cxx modules using `-fno-cxx-modules`. This
resolves the include/compilation issues.
Related discussion:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116827https://github.com/llvm/llvm-project/pull/150349https://github.com/llvm/llvm-project/issues/57432
The per-host flag variables hold platform-specific defaults that are ignored
when flag environment variables are set, so it was wrong for them to contain
-std options relied on by package definitions.
Additionally, these flags (-pipe and -std=xx) will no longer be passed into
the CMake build, meaning less duplication in the build summary.
Pulled out of #31920.
Similar to #31840, currently our Linux toolchain file contains:
```bash
set(CMAKE_AR "aarch64-linux-gnu-ar")
set(CMAKE_RANLIB "aarch64-linux-gnu-ranlib")
set(CMAKE_STRIP "aarch64-linux-gnu-strip")
set(CMAKE_OBJCOPY "aarch64-linux-gnu-objcopy")
set(CMAKE_OBJDUMP "")
```
`objdump` is currently only used for the macOS cross build, where it's
`llvm-objdump`, but we should be consistent in producing a toolchain
file that points to actual tools, rather than leaving variables unset.
Our CMake toolchain for a Darwin cross build currently contains:
```bash
set(CMAKE_AR "/usr/bin/llvm-ar")
set(CMAKE_RANLIB "/usr/bin/llvm-ranlib")
set(CMAKE_STRIP "/usr/bin/llvm-strip")
set(CMAKE_OBJCOPY "arm64-apple-darwin-objcopy")
set(CMAKE_OBJDUMP "/usr/bin/llvm-objdump")
```
`objcopy` isn't currently used for the Darwin build (only for Linux and
splitting the debug symbols), but we shouldn't be producing a toolchain
file that refers to nonexistent tools.
Running Bitcoin Core on unsupported OSes may expose users to security
issues.
macOS Monterey 12 received its final security update (12.7.6) on July
2024. Apple classifies the hardware that can run macOS 12 at most as
"obsolete worldwide".
CMake parses some paths from the spec of the C compiler, assuming it
will be the linker, resulting in the link to end up with
`-L/usr/lib/gcc/x86_64-w64-mingw32/12-win32` on debian bookworm if both
-win32 and -posix variants are installed, and -win32 is the default
alternative.
This results in the wrong C++ library being linked, missing
std::threads::hardware_concurrency and other threading functions.
To fix this, use the -posix variant of gcc as well when available. This
fixes a regression compared to autotools, where this scenario worked.
Now that we use the native compiler, and have fixed Qt, and these vars
are unset it Guix, we can remove the unsetting from our compiler command
here.
Fixes#21552.
Whilst these remain aliases for each other, the later is preferred,
and I assume the former will be removed at some point in the future;
see: https://github.com/llvm/llvm-project/pull/95374.
5deb0b024e build, test, doc: Temporarily remove Android-related stuff (Hennadii Stepanov)
Pull request description:
Previously, our Android builds were geared towards generating APKs, which relied on Qt. However, after migrating to C++20, compiling for Android became unfeasible due to Qt 5.15's compatibility limitations with NDK only up to r25, which includes an outdated embedded libc++ (see https://github.com/bitcoin/bitcoin/issues/29360).
All removed stuff will be reinstated after migrating the build system to CMake and upgrading Qt to version 6.x.
This PR makes possible a clean migration to the CMake-based build system as it removes code, which is not used at this moment.
ACKs for top commit:
vasild:
ACK 5deb0b024e
fanquake:
ACK 5deb0b024e - given none of this is currently tested/wont compile. Can be revisted in future.
Tree-SHA512: 3bc2ccfe881e11cc1d78c27acd6f1d86cfba86821ef3bb5eca2e80d978fdfa13659ec82284dcaadc507e2394524dea91d4b8f81d0030c1cef9708df8be76bf07
Adjust the security check for:
ld64.lld: warning: Option `-allow_stack_execute' is not yet implemented.
ld64.lld: error: -fixup_chains is incompatible with -no_pie
and to account for the embedding of LLVMs version number.
Similar to libtool, (llvm-)otool only exists with a version suffix
on some systems (Ubuntu), which makes it annoying to use/find. Avoid
this, by switching to objdump. Which is a drop-in replacement.
This is related to #21778, and the switchover to using vanilla LLVM for
macOS.
Previously, our Android builds were geared towards generating APKs,
which relied on Qt. However, after migrating to C++20, compiling for
Android became unfeasible due to Qt 5.15's compatibility limitations
with NDK only up to r25, which includes an outdated embedded libc++.
All removed stuff will be reinstated after migrating the build system to
CMake and upgrading Qt to version 6.x."
This has outlived its usefulness, doesn't gel well with
newer compilers & `-flto` related options, i.e thin vs full, or `=auto`,
and having `-flto` as the only option means that sometimes this just
needs to be worked around, i.e in oss-fuzz:
https://github.com/google/oss-fuzz/blob/master/projects/bitcoin-core/build.sh.
While it was convenient when `-flto` was newer, support for `-flto` is now
in all compilers we use, and there's also no-longer any real need
for us to treat `-flto` different to any other optimization option.
Remove it, to remove build complexity, and so there's no need
to port a similar option to CMake.
Note that the LTO option remains in depends, because we still a way to
build packages that have LTO specific patches/options.
If we decide to merge this, I'll follow up downstream in oss-fuzz first,
to make sure we don't break the build.
```bash
-platform_version platform min_version sdk_version
This is set to indicate the platform, oldest supported version
of that platform that output is to be used on, and the SDK that
the output was built against.
```
`_LIBCPP_ENABLE_ASSERTIONS` is deprecated, and will be removed. [See (from libc++ __config in main)](b57df9fe9a/libcxx/include/__config (L205-L209)):
> TODO(hardening): remove this in LLVM 19.
> This is for backward compatibility -- make enabling `_LIBCPP_ENABLE_ASSERTIONS` (which predates hardening modes)
> equivalent to setting the safe mode.
> ifdef _LIBCPP_ENABLE_ASSERTIONS
> warning "_LIBCPP_ENABLE_ASSERTIONS is deprecated, please use _LIBCPP_ENABLE_SAFE_MODE instead."
From LLVM 17, `_LIBCPP_ENABLE_DEBUG_MODE` can be used instead.
See https://libcxx.llvm.org/Hardening.html.
Related to #28476.
Use of -stdlib++-isystem gets rid of any system c++ header include paths and
negates the need for this option. In newer versions of clangs the combo
produces a warning.
It was deprecated in LLVM 15, turned into a compile-time error in LLVM 16:
```bash
In file included from /usr/lib/llvm-16/bin/../include/c++/v1/cassert:19:
/usr/lib/llvm-16/bin/../include/c++/v1/__assert:22:5: error: "Defining _LIBCPP_DEBUG is not supported anymore.
Please use _LIBCPP_ENABLE_DEBUG_MODE instead."
^
1 error generated.
```
and has been removed entirely in LLVM 17 (main),
ff573a42cd.
Building libc++ in debug mode, will also automatically set
`_LIBCPP_ENABLE_DEBUG_MODE` (the new define), so adding it to depends
doesn't seem useful, and would just result in redefinition errors.
I'm wondering if as a followup, we could enable a DEBUG build of libc++
in our MSAN CI job?
Somewhat related to https://github.com/google/oss-fuzz/pull/9828, where
it looks like we'll have to sort out getting a DEBUG build of LLVM.
a3a2bd9e8a ci: Drop no longer needed package-specific flags (Hennadii Stepanov)
071eef1e97 build: Propagate user-defined flags to host packages (Hennadii Stepanov)
Pull request description:
On master (4f8b1f8759) `{CPP,C,CXX,LD}FLAGS` that are specified in the command line are not propagated to packages:
```
$ make --no-print-directory -C depends print-libevent_cxxflags CXXFLAGS=-some-fancy-flag
libevent_cxxflags=-pipe -O2
```
This PR:
- propagates `{CPP,C,CXX,LD}FLAGS` to host packages:
```
$ make --no-print-directory -C depends print-libevent_cxxflags CXXFLAGS=-some-fancy-flag
libevent_cxxflags= -some-fancy-flag
```
- does not propagate `{CPP,C,CXX,LD}FLAGS` to native packages:
```
$ make --no-print-directory -C depends print-native_b2_cxxflags CXXFLAGS=-some-fancy-flag
native_b2_cxxflags=
```
- actually addresses the https://github.com/bitcoin/bitcoin/pull/23551#issuecomment-973896518
ACKs for top commit:
TheCharlatan:
Code review ACK a3a2bd9e8a
Tree-SHA512: 243d6b1b0e9c5de46debc36de62a77b6b4d6f638940fd530040c219956ec624e321b0c25290fed164e3a8c88befa7b97b20f765d7b9a428c269b3720f21da099
Use lto.h from clang+llvm not libtapi. The later is older,
and comes bundled with the libtapi repo.
Copy libLTO.so when building with FORCE_USE_SYSTEM_CLANG.
