fix problems exposed by eschewing static_get_completion_signatures and static_connect members (#1720)

Author: ericniebler

.clangd

@@ -80,6 +80,7 @@ CompileFlags:
     - -profiler
     - -ignore-system
     - "-fconcepts-diagnostics-depth=*"
+    - "-gopt"
 Diagnostics:
   Suppress:
     - "variadic_device_fn"

.vscode/launch.json

@@ -74,15 +74,15 @@
             "args": "-v normal ${input:CUDA_TEST_TAGS}",
         },
         {
-            "name": "maxwell_gpu_s (cuda-gdb)",
+            "name": "maxwell_cpu_st (cuda-gdb)",
             "type": "cuda-gdb",
             "request": "launch",
             "stopAtEntry": false,
             "breakOnLaunch": false,
             "internalConsoleOptions": "openOnFirstSessionStart",
-            "program": "${command:cmake.buildDirectory}/examples/nvexec/maxwell_gpu_s",
+            "program": "${command:cmake.buildDirectory}/examples/nvexec/maxwell_cpu_st",
             "cwd": "${command:cmake.buildDirectory}",
-            "args": ["--iterations=1000", "--N=512", "--run-cuda", "--run-stdpar", "--run-stream-scheduler"],
+            "args": ["--iterations=1000", "--N=512", "--run-cpp", "--run-inline-scheduler"],
         },
     ],
     "inputs": [

examples/nvexec/maxwell/common.cuh

@@ -56,7 +56,6 @@ struct deleter_t {
 };
 
 template <class T>
-STDEXEC_ATTRIBUTE(host, device)
 inline auto allocate_on(bool gpu, std::size_t elements = 1) -> std::unique_ptr<T, deleter_t> {
   T *ptr{};
 

examples/nvexec/maxwell_cpu_mt.cu

@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+// This file causes clangd to crash during parsing
+#if !defined(STDEXEC_CLANGD_INVOKED)
+
 #include "maxwell/snr.cuh"
 #include "maxwell/std.cuh"
 #include "maxwell/stdpar.cuh" // IWYU pragma: keep
@@ -85,3 +88,5 @@ auto main(int argc, char *argv[]) -> int {
   }
 #endif
 }
+
+#endif // !defined(STDEXEC_CLANGD_INVOKED)

examples/nvexec/maxwell_cpu_st.cu

@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+// This file causes clangd to crash during parsing
+#if !defined(STDEXEC_CLANGD_INVOKED)
+
 #include "maxwell/snr.cuh"
 #include "maxwell/cpp.cuh"
 
@@ -65,3 +68,5 @@ auto main(int argc, char *argv[]) -> int {
     run_cpp(dt, write_vtk, n_iterations, grid, "CPU (cpp)");
   }
 }
+
+#endif // !defined(STDEXEC_CLANGD_INVOKED)

examples/nvexec/maxwell_distributed.cu

@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+// This file causes clangd to crash during parsing
+#if !defined(STDEXEC_CLANGD_INVOKED)
+
 #include "maxwell/snr.cuh"           // IWYU pragma: keep
 #include "nvexec/stream_context.cuh" // IWYU pragma: keep
 
@@ -512,7 +515,7 @@ auto main(int argc, char *argv[]) -> int {
   }
 
   write();
-#  else
+#  else // ^^ defined(OVERLAP) ^^ // vv !defined(OVERLAP) vv
   for (std::size_t compute_step = 0; compute_step < n_iterations; compute_step++) {
     auto compute_h =
       ex::just()
@@ -531,7 +534,7 @@ auto main(int argc, char *argv[]) -> int {
   }
 
   write();
-#  endif
+#  endif // defined(OVERLAP)
 
   MPI_Barrier(MPI_COMM_WORLD);
   const auto end = std::chrono::system_clock::now();
@@ -550,3 +553,5 @@ auto main(int argc, char *argv[]) -> int {
   MPI_Finalize();
 }
 #endif
+
+#endif // !defined(STDEXEC_CLANGD_INVOKED)

examples/nvexec/maxwell_gpu_m.cu

@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+// This file causes clangd to crash during parsing
+#if !defined(STDEXEC_CLANGD_INVOKED)
+
 #include "maxwell/snr.cuh"
 
 auto main(int argc, char *argv[]) -> int {
@@ -52,3 +55,5 @@ auto main(int argc, char *argv[]) -> int {
   nvexec::multi_gpu_stream_context stream_context{};
   run_snr_on("GPU (multi-GPU)", stream_context.get_scheduler());
 }
+
+#endif // !defined(STDEXEC_CLANGD_INVOKED)

examples/nvexec/maxwell_gpu_s.cu

@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+// This file causes clangd to crash during parsing
+#if !defined(STDEXEC_CLANGD_INVOKED)
+
 #include "maxwell/snr.cuh"
 #include "maxwell/stdpar.cuh" // IWYU pragma: keep
 #include "maxwell/cuda.cuh"
@@ -79,5 +82,7 @@ auto main(int argc, char *argv[]) -> int {
 
     run_stdpar(dt, write_vtk, n_iterations, grid, stdexec::par_unseq, method);
   }
-#endif
+#endif // STDEXEC_HAS_PARALLEL_ALGORITHMS()
 }
+
+#endif // !defined(STDEXEC_CLANGD_INVOKED)

include/exec/sequence.hpp

@@ -186,15 +186,15 @@ namespace exec {
         stdexec::__minvoke<stdexec::__mfold_left<void, stdexec::__q2<_fold_fn>>, Sndrs...>;
     };
 
-    template <class... Sndrs>
-    struct _sndr {
+    template <class Sndr0, class... Sndrs>
+    struct _sndr<Sndr0, Sndrs...> {
       using sender_concept = stdexec::sender_t;
 
       template <class... Env>
-      using _completions_t = stdexec::__minvoke<_completions<Env...>, Sndrs...>;
+      using _completions_t = stdexec::__minvoke<_completions<Env...>, Sndr0, Sndrs...>;
 
       template <class Self, class... Env>
-        requires(stdexec::__decay_copyable<stdexec::__copy_cvref_t<Self, Sndrs>> && ...)
+        requires stdexec::__decay_copyable<Self>
       STDEXEC_ATTRIBUTE(host, device)
       STDEXEC_EXPLICIT_THIS_BEGIN(auto get_completion_signatures)(this Self&&, Env&&...)
         -> _completions_t<Env...> {
@@ -203,17 +203,17 @@ namespace exec {
       STDEXEC_EXPLICIT_THIS_END(get_completion_signatures)
 
       template <class Self, class Rcvr>
-        requires(stdexec::__decay_copyable<stdexec::__copy_cvref_t<Self, Sndrs>> && ...)
+        requires stdexec::__decay_copyable<Self>
       STDEXEC_ATTRIBUTE(host, device)
       STDEXEC_EXPLICIT_THIS_BEGIN(auto connect)(this Self&& self, Rcvr rcvr) {
-        return _opstate<Rcvr, Sndrs...>{
+        return _opstate<Rcvr, stdexec::__copy_cvref_t<Self, Sndr0>, Sndrs...>{
           static_cast<Rcvr&&>(rcvr), static_cast<Self&&>(self)._sndrs};
       }
       STDEXEC_EXPLICIT_THIS_END(connect)
 
       STDEXEC_ATTRIBUTE(no_unique_address, maybe_unused) sequence_t _tag;
       STDEXEC_ATTRIBUTE(no_unique_address, maybe_unused) stdexec::__ignore _ignore;
-      stdexec::__tuple<Sndrs...> _sndrs;
+      stdexec::__tuple<Sndr0, Sndrs...> _sndrs;
     };
 
     template <class Sndr>

include/exec/sequence/merge_each.hpp

@@ -32,7 +32,6 @@
 #include "../../stdexec/__detail/__senders_core.hpp"
 #include "../../stdexec/__detail/__stop_token.hpp"
 #include "../../stdexec/__detail/__transform_completion_signatures.hpp"
-#include "../../stdexec/__detail/__unstoppable.hpp"
 #include "../../stdexec/__detail/__variant.hpp"
 
 #include <atomic>
@@ -240,7 +239,7 @@ namespace exec {
 
         __operation_base_interface_t* __op_;
 
-        template <std::same_as<__t> _Self, class... _Env>
+        template <class _Self, class... _Env>
         STDEXEC_EXPLICIT_THIS_BEGIN(
           auto get_completion_signatures)(this _Self&&, _Env&&...) noexcept
           -> stdexec::__mapply<
@@ -252,8 +251,9 @@ namespace exec {
           > {
           return {};
         }
+        STDEXEC_EXPLICIT_THIS_END(get_completion_signatures)
 
-        template <std::same_as<__t> _Self, receiver _ErrorReceiver>
+        template <class _Self, receiver _ErrorReceiver>
         STDEXEC_EXPLICIT_THIS_BEGIN(auto connect)(this _Self&& __self, _ErrorReceiver&& __rcvr)
           noexcept(__nothrow_move_constructible<_ErrorReceiver>)
             -> __error_op_t<stdexec::__id<_ErrorReceiver>> {
@@ -607,7 +607,7 @@ namespace exec {
         _NestedValueSender __nested_value_;
         __operation_base_interface_t* __op_;
 
-        template <std::same_as<__t> _Self, class... _Env>
+        template <class _Self, class... _Env>
         STDEXEC_EXPLICIT_THIS_BEGIN(
           auto get_completion_signatures)(this _Self&&, _Env&&...) noexcept
           -> stdexec::transform_completion_signatures<
@@ -618,8 +618,9 @@ namespace exec {
           > {
           return {};
         }
+        STDEXEC_EXPLICIT_THIS_END(get_completion_signatures)
 
-        template <std::same_as<__t> _Self, receiver _NestedValueReceiver>
+        template <class _Self, receiver _NestedValueReceiver>
         STDEXEC_EXPLICIT_THIS_BEGIN(
           auto connect)(this _Self&& __self, _NestedValueReceiver&& __rcvr)
           noexcept(__nothrow_constructible_from<
@@ -1072,14 +1073,16 @@ namespace exec {
         _OperationBase* __op_;
         _NestedSequenceSender __nested_sequence_;
 
-        template <std::same_as<__t> _Self, class... _Env>
+        template <class _Self, class... _Env>
         STDEXEC_EXPLICIT_THIS_BEGIN(
           auto get_completion_signatures)(this _Self&&, _Env&&...) noexcept
           -> stdexec::completion_signatures<stdexec::set_value_t(), stdexec::set_stopped_t()> {
           return {};
         }
+        STDEXEC_EXPLICIT_THIS_END(get_completion_signatures)
 
-        template <std::same_as<__t> _Self, receiver _NextReceiver>
+
+        template <class _Self, receiver _NextReceiver>
         STDEXEC_EXPLICIT_THIS_BEGIN(auto connect)(this _Self&& __self, _NextReceiver&& __rcvr)
           noexcept(__nothrow_constructible_from<
                    __next_sequence_op_t<stdexec::__id<_NextReceiver>>,