storeChunk: Workaround for bug with parallel flushing (#5510)

* Little fix for parallel flushing

* Workaround for openPMD/openPMD-api#1794

* Add the workaround also for recent releases

Author: franzpoeschel

include/picongpu/plugins/openPMD/writer/ParticleAttribute.hpp

@@ -158,6 +158,20 @@ namespace picongpu
                 if(elements == 0)
                 {
                     // accumulateWrittenBytes += 0;
+
+                    // Workaround for this bug: https://github.com/openPMD/openPMD-api/pull/1794
+                    // and also this one for good measure https://github.com/openPMD/openPMD-api/pull/1862
+                    // In the affected versions of the openPMD-api, Span-based storeChunk must be
+                    // treated as an MPI-collective call.
+                    for(uint32_t d = 0; d < components; d++)
+                    {
+                        ::openPMD::RecordComponent recordComponent
+                            = components > 1 ? record[name_lookup[d]] : record[::openPMD::MeshRecordComponent::SCALAR];
+
+                        recordComponent.storeChunk<ComponentType>(
+                            ::openPMD::Offset{globalOffset},
+                            ::openPMD::Extent{elements});
+                    }
                     return;
                 }