Changing how execution parallelism is handled, so that different

executors can implement their own idea of parallelism.

tests/unit_threadpool.cpp

@@ -38,4 +38,25 @@ TEST_CASE("threadpool", "[threadpool]") {
         for (auto &r: res) r.get();
         REQUIRE(cnt == 100);
     }
+
+    SECTION("parallel") {
+        std::vector<std::future<void>> res;
+        using namespace std::chrono_literals;
+        std::atomic<uint32_t> maxCnt{0};
+        for (size_t i = 0; i < 100; ++i)
+            res.push_back(tp.addTask([&cnt,&maxCnt, i]() {
+                auto delay = 20ms;
+                uint32_t current = cnt.fetch_add(1);
+                delay += i * 1ms;
+                std::this_thread::sleep_for(delay);
+                if (current > maxCnt) {
+                    maxCnt = current;
+                }
+                cnt--;
+                return;
+            }));
+        for (auto &r: res) r.get();
+        REQUIRE(maxCnt > 1);
+    }
+
 }