Building with clang's libc++ fails #569
Description
PR #549 modified Nighthawk's use of MonotonicTime to SystemTime, which causes a failure to build with clang's/llvm's libc++ (--config=libc++), and gives the following error:
ERROR: /home/.../git/nighthawk/source/client/BUILD:11:17: C++ compilation of rule '//source/client:nighthawk_client_lib' failed (Exit 1) clang failed: error executing command /home/.../clang10/bin/clang -U_FORTIFY_SOURCE -fstack-protector -Wall -Wthread-safety -Wself-assign -fcolor-diagnostics -fno-omit-frame-pointer '-stdlib=libc++' -MD -MF ... (remaining 765 argument(s) skipped)
Use --sandbox_debug to see verbose messages from the sandbox
In file included from source/client/process_impl.cc:1:
In file included from bazel-out/k8-fastbuild/bin/source/client/_virtual_includes/nighthawk_client_lib/client/process_impl.h:3:
In file included from /home/.../clang10/bin/../include/c++/v1/map:479:
In file included from /home/.../clang10/bin/../include/c++/v1/__tree:15:
/home/.../clang10/bin/../include/c++/v1/memory:3028:32: error: no matching constructor for initialization of 'Nighthawk::Client::ClientWorkerImpl'
return unique_ptr<_Tp>(new _Tp(_VSTD::forward<_Args>(__args)...));
^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
source/client/process_impl.cc:188:29: note: in instantiation of function template specialization 'std::__1::make_unique<Nighthawk::Client::ClientWorkerImpl, Envoy::Api::Api &, Envoy::ThreadLocal::InstanceImpl &, std::__1::unique_ptr<Envoy::Upstream::ClusterManager, std::__1::default_delete<Envoy::Upstream::ClusterManager> > &, const Nighthawk::Client::BenchmarkClientFactoryImpl &, const Nighthawk::Client::TerminationPredicateFactoryImpl &, const Nighthawk::Client::SequencerFactoryImpl &, const Nighthawk::Client::RequestSourceFactoryImpl &, Envoy::Stats::ThreadLocalStoreImpl &, int &, std::__1::chrono::time_point<std::__1::chrono::system_clock, std::__1::chrono::duration<long long, std::__1::ratio<1, 1000000000> > >, std::__1::shared_ptr<Envoy::Tracing::HttpTracer> &, Nighthawk::Client::ClientWorkerImpl::HardCodedWarmupStyle>' requested here
workers_.push_back(std::make_unique<ClientWorkerImpl>(
^
bazel-out/k8-fastbuild/bin/source/client/_virtual_includes/nighthawk_client_lib/client/client_worker_impl.h:29:3: note: candidate constructor not viable: no known conversion from 'time_point<[...], duration<[...], ratio<[...], 1000000000>>>' to 'const time_point<[...], duration<[...], ratio<[...], 1000000>>>' for 10th argument
ClientWorkerImpl(Envoy::Api::Api& api, Envoy::ThreadLocal::Instance& tls,
^
bazel-out/k8-fastbuild/bin/source/client/_virtual_includes/nighthawk_client_lib/client/client_worker_impl.h:25:7: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 12 were provided
class ClientWorkerImpl : public WorkerImpl, virtual public ClientWorker {
^
1 error generated.
Target //source/client:nighthawk_client_lib failed to build
After executing:
bazel build //source/client:nighthawk_client_lib --config=libc++
The cause seems to be the different resolution of system_clock by llvm and gcc:
- llvm uses a milliseconds resolution;
- gcc uses a nanoseconds resolution.
and the nighthawk code requires a nanoseconds resolution.
Possible ways to solve:
- Revert, and change back to
MonotonicTime. - Use
MonotonicTimein this part of the code. - Change the required resolution in this code to milliseconds instead of nanoseconds.
- Change Envoy's definition of SystemTime to explicitly use a nanoseconds, i.e.:
using SystemTime = std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds>;, and update the necessary places in Envoy's code to correctly cast when system_clock is used (std::chrono::time_point_cast<std::chrono::system_clock::duration>(time)).
@oschaaf Am I missing anything? What's your opinion about this?