Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Update copyright lines for 2023.
[simgrid.git] / src / s4u / s4u_Comm.cpp
index f348f68..ad42fb4 100644 (file)
-/* Copyright (c) 2006-2020. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2006-2023. The SimGrid Team. All rights reserved.          */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
-#include "src/msg/msg_private.hpp"
-#include "xbt/log.h"
-
-#include "simgrid/Exception.hpp"
-#include "simgrid/s4u/Comm.hpp"
-#include "simgrid/s4u/Mailbox.hpp"
-
+#include <cmath>
+#include <simgrid/Exception.hpp>
 #include <simgrid/comm.h>
+#include <simgrid/s4u/Comm.hpp>
+#include <simgrid/s4u/Engine.hpp>
+#include <simgrid/s4u/Mailbox.hpp>
+
+#include "mc/mc.h"
+#include "src/kernel/activity/CommImpl.hpp"
+#include "src/kernel/actor/ActorImpl.hpp"
+#include "src/kernel/actor/SimcallObserver.hpp"
+#include "src/mc/mc_replay.hpp"
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
 
-namespace simgrid {
-namespace s4u {
-xbt::signal<void(Actor const&)> Comm::on_sender_start;
-xbt::signal<void(Actor const&)> Comm::on_receiver_start;
-xbt::signal<void(Actor const&)> Comm::on_completion;
+namespace simgrid::s4u {
+xbt::signal<void(Comm const&)> Comm::on_send;
+xbt::signal<void(Comm const&)> Comm::on_recv;
+
+CommPtr Comm::set_copy_data_callback(const std::function<void(kernel::activity::CommImpl*, void*, size_t)>& callback)
+{
+  copy_data_function_ = callback;
+  return this;
+}
+
+void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
+{
+  XBT_DEBUG("Copy the data over");
+  memcpy(comm->dst_buff_, buff, buff_size);
+  if (comm->is_detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
+                             // original buffer available to the application ASAP
+    xbt_free(buff);
+    comm->src_buff_ = nullptr;
+  }
+}
+
+void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
+{
+  xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
+  *(void**)(comm->dst_buff_) = buff;
+}
 
 Comm::~Comm()
 {
   if (state_ == State::STARTED && not detached_ &&
-      (pimpl_ == nullptr || pimpl_->state_ == kernel::activity::State::RUNNING)) {
-    XBT_INFO("Comm %p freed before its completion. Detached: %d, State: %d", this, detached_, (int)state_);
+      (pimpl_ == nullptr || pimpl_->get_state() == kernel::activity::State::RUNNING)) {
+    XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str());
     if (pimpl_ != nullptr)
-      XBT_INFO("pimpl_->state: %d", static_cast<int>(pimpl_->state_));
+      XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str());
     else
       XBT_INFO("pimpl_ is null");
     xbt_backtrace_display_current();
   }
 }
 
-int Comm::wait_any_for(const std::vector<CommPtr>* comms, double timeout)
+void Comm::send(kernel::actor::ActorImpl* sender, const Mailbox* mbox, double task_size, double rate, void* src_buff,
+                size_t src_buff_size,
+                const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
+                const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
+                void* data, double timeout)
 {
-  std::unique_ptr<kernel::activity::CommImpl* []> rcomms(new kernel::activity::CommImpl*[comms->size()]);
-  std::transform(begin(*comms), end(*comms), rcomms.get(),
-                 [](const CommPtr& comm) { return static_cast<kernel::activity::CommImpl*>(comm->pimpl_.get()); });
-  int changed_pos = simcall_comm_waitany(rcomms.get(), comms->size(), timeout);
-  if (changed_pos != -1)
-    comms->at(changed_pos)->release_dependencies();
-  return changed_pos;
+  /* checking for infinite values */
+  xbt_assert(std::isfinite(task_size), "task_size is not finite!");
+  xbt_assert(std::isfinite(rate), "rate is not finite!");
+  xbt_assert(std::isfinite(timeout), "timeout is not finite!");
+
+  xbt_assert(mbox, "No rendez-vous point defined for send");
+
+  if (MC_is_active() || MC_record_replay_is_active()) {
+    /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
+    simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
+
+    simgrid::kernel::actor::CommIsendSimcall send_observer{
+        sender,  mbox->get_impl(), task_size, rate, static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
+        nullptr, copy_data_fun,    data,      false};
+    comm = simgrid::kernel::actor::simcall_answered(
+        [&send_observer] { return simgrid::kernel::activity::CommImpl::isend(&send_observer); }, &send_observer);
+
+    if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{sender, comm.get(), timeout};
+        simgrid::kernel::actor::simcall_blocking(
+            [&wait_observer] {
+              wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
+            },
+            &wait_observer)) {
+      throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
+    }
+    comm = nullptr;
+  } else {
+    simgrid::kernel::actor::CommIsendSimcall observer(sender, mbox->get_impl(), task_size, rate,
+                                                      static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
+                                                      nullptr, copy_data_fun, data, false);
+    simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
+      simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer);
+      comm->wait_for(observer.get_issuer(), timeout);
+    });
+  }
 }
 
-void Comm::wait_all(const std::vector<CommPtr>* comms)
+void Comm::recv(kernel::actor::ActorImpl* receiver, const Mailbox* mbox, void* dst_buff, size_t* dst_buff_size,
+                const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
+                const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
+                void* data, double timeout, double rate)
 {
-  // TODO: this should be a simcall or something
-  // TODO: we are missing a version with timeout
-  for (CommPtr comm : *comms)
-    comm->wait();
+  xbt_assert(std::isfinite(timeout), "timeout is not finite!");
+  xbt_assert(mbox, "No rendez-vous point defined for recv");
+
+  if (MC_is_active() || MC_record_replay_is_active()) {
+    /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
+    simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
+
+    simgrid::kernel::actor::CommIrecvSimcall observer{receiver,
+                                                      mbox->get_impl(),
+                                                      static_cast<unsigned char*>(dst_buff),
+                                                      dst_buff_size,
+                                                      match_fun,
+                                                      copy_data_fun,
+                                                      data,
+                                                      rate};
+    comm = simgrid::kernel::actor::simcall_answered(
+        [&observer] { return simgrid::kernel::activity::CommImpl::irecv(&observer); }, &observer);
+
+    if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{receiver, comm.get(), timeout};
+        simgrid::kernel::actor::simcall_blocking(
+            [&wait_observer] {
+              wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
+            },
+            &wait_observer)) {
+      throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
+    }
+    comm = nullptr;
+  } else {
+    simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox->get_impl(), static_cast<unsigned char*>(dst_buff),
+                                                      dst_buff_size, match_fun, copy_data_fun, data, rate);
+    simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
+      simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer);
+      comm->wait_for(observer.get_issuer(), timeout);
+    });
+  }
+}
+
+CommPtr Comm::sendto_init()
+{
+  CommPtr res(new Comm());
+  res->pimpl_ = kernel::activity::CommImplPtr(new kernel::activity::CommImpl());
+  boost::static_pointer_cast<kernel::activity::CommImpl>(res->pimpl_)->detach();
+  res->sender_ = kernel::actor::ActorImpl::self();
+  return res;
+}
+
+CommPtr Comm::sendto_init(Host* from, Host* to)
+{
+  auto res = Comm::sendto_init()->set_source(from)->set_destination(to);
+  res->set_state(State::STARTING);
+  return res;
+}
+
+CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
+{
+  return Comm::sendto_init()->set_payload_size(simulated_size_in_bytes)->set_source(from)->set_destination(to);
+}
+
+void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
+{
+  sendto_async(from, to, simulated_size_in_bytes)->wait();
+}
+
+CommPtr Comm::set_source(Host* from)
+{
+  xbt_assert(state_ == State::INITED || state_ == State::STARTING,
+             "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
+  boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_source(from);
+  // Setting 'source' may allow to start the activity, let's try
+  if (state_ == State::STARTING && remains_ <= 0)
+    XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
+  else
+    vetoable_start();
+
+  return this;
+}
+Host* Comm::get_source() const
+{
+  return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_source() : nullptr;
+}
+
+CommPtr Comm::set_destination(Host* to)
+{
+  xbt_assert(state_ == State::INITED || state_ == State::STARTING,
+             "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
+  boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_destination(to);
+  // Setting 'destination' may allow to start the activity, let's try
+  if (state_ == State::STARTING && remains_ <= 0)
+    XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
+  else
+    vetoable_start();
+
+  return this;
+}
+
+Host* Comm::get_destination() const
+{
+  return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_destination() : nullptr;
 }
 
 CommPtr Comm::set_rate(double rate)
@@ -60,6 +214,14 @@ CommPtr Comm::set_rate(double rate)
   return this;
 }
 
+CommPtr Comm::set_mailbox(Mailbox* mailbox)
+{
+  xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
+             __FUNCTION__);
+  mailbox_ = mailbox;
+  return this;
+}
+
 CommPtr Comm::set_src_data(void* buff)
 {
   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
@@ -97,11 +259,6 @@ CommPtr Comm::set_dst_data(void** buff)
   return this;
 }
 
-size_t Comm::get_dst_data_size() const
-{
-  xbt_assert(state_ == State::FINISHED, "You cannot use %s before your communication terminated", __FUNCTION__);
-  return dst_buff_size_;
-}
 CommPtr Comm::set_dst_data(void** buff, size_t size)
 {
   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
@@ -113,39 +270,102 @@ CommPtr Comm::set_dst_data(void** buff, size_t size)
   return this;
 }
 
-CommPtr Comm::set_tracing_category(const std::string& category)
+CommPtr Comm::set_payload_size(uint64_t bytes)
 {
-  xbt_assert(state_ == State::INITED, "Cannot change the tracing category of an exec after its start");
-  tracing_category_ = category;
+  set_remaining(bytes);
+  if (pimpl_) {
+    boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_size(bytes);
+  }
   return this;
 }
 
+Actor* Comm::get_sender() const
+{
+  kernel::actor::ActorImplPtr sender = nullptr;
+  if (pimpl_)
+    sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
+  return sender ? sender->get_ciface() : nullptr;
+}
+
+bool Comm::is_assigned() const
+{
+  return (pimpl_ && boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->is_assigned()) ||
+         mailbox_ != nullptr;
+}
+
 Comm* Comm::start()
 {
   xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
              "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
-
-  if (src_buff_ != nullptr) { // Sender side
-    on_sender_start(*Actor::self());
-    pimpl_ = simcall_comm_isend(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_,
-                                clean_fun_, copy_data_function_, get_user_data(), detached_);
+  if (get_source() != nullptr || get_destination() != nullptr) {
+    xbt_assert(is_assigned(), "When either from_ or to_ is specified, both must be.");
+    xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
+               "Direct host-to-host communications cannot carry any data.");
+    XBT_DEBUG("host-to-host Comm. Pimpl already created and set, just start it.");
+    kernel::actor::simcall_answered([this] {
+      pimpl_->set_state(kernel::activity::State::READY);
+      boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->start();
+    });
+  } else if (src_buff_ != nullptr) { // Sender side
+    on_send(*this);
+    kernel::actor::CommIsendSimcall observer{sender_,
+                                             mailbox_->get_impl(),
+                                             remains_,
+                                             rate_,
+                                             static_cast<unsigned char*>(src_buff_),
+                                             src_buff_size_,
+                                             match_fun_,
+                                             clean_fun_,
+                                             copy_data_function_,
+                                             get_data<void>(),
+                                             detached_};
+    pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::isend(&observer); },
+                                             &observer);
   } else if (dst_buff_ != nullptr) { // Receiver side
     xbt_assert(not detached_, "Receive cannot be detached");
-    on_receiver_start(*Actor::self());
-    pimpl_ = simcall_comm_irecv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_,
-                                copy_data_function_, get_user_data(), rate_);
-
+    on_recv(*this);
+    kernel::actor::CommIrecvSimcall observer{receiver_,
+                                             mailbox_->get_impl(),
+                                             static_cast<unsigned char*>(dst_buff_),
+                                             &dst_buff_size_,
+                                             match_fun_,
+                                             copy_data_function_,
+                                             get_data<void>(),
+                                             rate_};
+    pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::irecv(&observer); },
+                                             &observer);
   } else {
     xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
   }
+
+  if (suspended_)
+    pimpl_->suspend();
+
+  if (not detached_) {
+    pimpl_->set_iface(this);
+    pimpl_->set_actor(sender_);
+  }
+
   state_ = State::STARTED;
   return this;
 }
 
-/** @brief Block the calling actor until the communication is finished */
-Comm* Comm::wait()
+Comm* Comm::detach()
+{
+  xbt_assert(state_ == State::INITED || state_ == State::STARTING,
+             "You cannot use %s() once your communication is %s (not implemented)", __FUNCTION__, get_state_str());
+  xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
+  detached_ = true;
+  vetoable_start();
+  return this;
+}
+
+ssize_t Comm::test_any(const std::vector<CommPtr>& comms)
 {
-  return this->wait_for(-1);
+  std::vector<ActivityPtr> activities;
+  for (const auto& comm : comms)
+    activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
+  return Activity::test_any(activities);
 }
 
 /** @brief Block the calling actor until the communication is finished, or until timeout
@@ -156,31 +376,42 @@ Comm* Comm::wait()
  *                Negative values denote infinite wait times. 0 as a timeout returns immediately. */
 Comm* Comm::wait_for(double timeout)
 {
+  XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
+  kernel::actor::ActorImpl* issuer = nullptr;
   switch (state_) {
     case State::FINISHED:
       break;
-
+    case State::FAILED:
+      throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
     case State::INITED:
-    case State::STARTING: // It's not started yet. Do it in one simcall
-      if (src_buff_ != nullptr) {
-        on_sender_start(*Actor::self());
-        simcall_comm_send(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_,
-                          copy_data_function_, get_user_data(), timeout);
+    case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
+      if (get_source() != nullptr || get_destination() != nullptr) {
+        return vetoable_start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
+      } else if (src_buff_ != nullptr) {
+        on_send(*this);
+        send(sender_, mailbox_, remains_, rate_, src_buff_, src_buff_size_, match_fun_, copy_data_function_,
+             get_data<void>(), timeout);
 
       } else { // Receiver
-        on_receiver_start(*Actor::self());
-        simcall_comm_recv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_,
-                          get_user_data(), timeout, rate_);
+        on_recv(*this);
+        recv(receiver_, mailbox_, dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_data<void>(),
+             timeout, rate_);
       }
-      state_ = State::FINISHED;
-      this->release_dependencies();
       break;
-
     case State::STARTED:
-      simcall_comm_wait(get_impl(), timeout);
-      on_completion(*Actor::self());
-      state_ = State::FINISHED;
-      this->release_dependencies();
+      try {
+        issuer = kernel::actor::ActorImpl::self();
+        kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout};
+        if (kernel::actor::simcall_blocking(
+                [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
+                &observer)) {
+          throw TimeoutException(XBT_THROW_POINT, "Timeouted");
+        }
+      } catch (const NetworkFailureException& e) {
+        issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
+        complete(State::FAILED);
+        e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
+      }
       break;
 
     case State::CANCELED:
@@ -189,77 +420,68 @@ Comm* Comm::wait_for(double timeout)
     default:
       THROW_IMPOSSIBLE;
   }
+  complete(State::FINISHED);
   return this;
 }
 
-int Comm::test_any(const std::vector<CommPtr>* comms)
+ssize_t Comm::wait_any_for(const std::vector<CommPtr>& comms, double timeout)
 {
-  std::unique_ptr<kernel::activity::CommImpl* []> rcomms(new kernel::activity::CommImpl*[comms->size()]);
-  std::transform(begin(*comms), end(*comms), rcomms.get(),
-                 [](const CommPtr& comm) { return static_cast<kernel::activity::CommImpl*>(comm->pimpl_.get()); });
-  int changed_pos = simcall_comm_testany(rcomms.get(), comms->size());
-  if (changed_pos != -1)
-    comms->at(changed_pos)->release_dependencies();
+  std::vector<ActivityPtr> activities;
+  for (const auto& comm : comms)
+    activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
+  ssize_t changed_pos;
+  try {
+    changed_pos = Activity::wait_any_for(activities, timeout);
+  } catch (const NetworkFailureException& e) {
+    changed_pos = -1;
+    for (auto c : comms) {
+      if (c->pimpl_->get_state() == kernel::activity::State::FAILED) {
+        c->complete(State::FAILED);
+      }
+    }
+    e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
+  }
   return changed_pos;
 }
 
-Comm* Comm::detach()
+void Comm::wait_all(const std::vector<CommPtr>& comms)
 {
-  xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
-             __FUNCTION__);
-  xbt_assert(src_buff_ != nullptr && src_buff_size_ != 0, "You can only detach sends, not recvs");
-  detached_ = true;
-  vetoable_start();
-  return this;
-}
-
-Comm* Comm::cancel()
-{
-  kernel::actor::simcall([this] {
-    if (pimpl_)
-      boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->cancel();
-  });
-  state_ = State::CANCELED;
-  return this;
+  // TODO: this should be a simcall or something
+  for (auto& comm : comms)
+    comm->wait();
 }
 
-bool Comm::test()
+size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout)
 {
-  xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::STARTING ||
-             state_ == State::FINISHED);
-
-  if (state_ == State::FINISHED)
-    return true;
-
-  if (state_ == State::INITED || state_ == State::STARTING)
-    this->vetoable_start();
-
-  if (simcall_comm_test(get_impl())) {
-    state_ = State::FINISHED;
-    this->release_dependencies();
-    return true;
+  if (timeout < 0.0) {
+    wait_all(comms);
+    return comms.size();
   }
-  return false;
-}
-
-Mailbox* Comm::get_mailbox() const
-{
-  return mailbox_;
-}
 
-Actor* Comm::get_sender() const
-{
-  return sender_ ? sender_->ciface() : nullptr;
+  double deadline = Engine::get_clock() + timeout;
+  std::vector<CommPtr> waited_comm(1, nullptr);
+  for (size_t i = 0; i < comms.size(); i++) {
+    double wait_timeout = std::max(0.0, deadline - Engine::get_clock());
+    waited_comm[0]      = comms[i];
+    // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout
+    if (wait_any_for(waited_comm, wait_timeout) == -1) {
+      XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i);
+      return i;
+    }
+  }
+  return comms.size();
 }
-
-} // namespace s4u
-} // namespace simgrid
+} // namespace simgrid::s4u
 /* **************************** Public C interface *************************** */
 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
 {
-  comm = comm->detach(clean_function);
+  comm->detach(clean_function);
+  comm->unref();
+}
+void sg_comm_unref(sg_comm_t comm)
+{
+  comm->unref();
 }
-
 int sg_comm_test(sg_comm_t comm)
 {
   bool finished = comm->test();
@@ -270,19 +492,7 @@ int sg_comm_test(sg_comm_t comm)
 
 sg_error_t sg_comm_wait(sg_comm_t comm)
 {
-  sg_error_t status = SG_OK;
-
-  simgrid::s4u::CommPtr s4u_comm(comm, false);
-  try {
-    s4u_comm->wait_for(-1);
-  } catch (const simgrid::TimeoutException&) {
-    status = SG_ERROR_TIMEOUT;
-  } catch (const simgrid::CancelException&) {
-    status = SG_ERROR_CANCELED;
-  } catch (const simgrid::NetworkFailureException&) {
-    status = SG_ERROR_NETWORK;
-  }
-  return status;
+  return sg_comm_wait_for(comm, -1);
 }
 
 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
@@ -303,28 +513,36 @@ sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
 }
 
 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
+{
+  sg_comm_wait_all_for(comms, count, -1);
+}
+
+size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout)
 {
   std::vector<simgrid::s4u::CommPtr> s4u_comms;
-  for (unsigned int i = 0; i < count; i++)
+  for (size_t i = 0; i < count; i++)
     s4u_comms.emplace_back(comms[i], false);
 
-  simgrid::s4u::Comm::wait_all(&s4u_comms);
+  size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout);
+  for (size_t i = pos; i < count; i++)
+    s4u_comms[i]->add_ref();
+  return pos;
 }
 
-int sg_comm_wait_any(sg_comm_t* comms, size_t count)
+ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
 {
   return sg_comm_wait_any_for(comms, count, -1);
 }
 
-int sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
+ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
 {
   std::vector<simgrid::s4u::CommPtr> s4u_comms;
-  for (unsigned int i = 0; i < count; i++)
+  for (size_t i = 0; i < count; i++)
     s4u_comms.emplace_back(comms[i], false);
 
-  int pos = simgrid::s4u::Comm::wait_any_for(&s4u_comms, timeout);
-  for (unsigned i = 0; i < count; i++) {
-    if (pos != -1 && static_cast<unsigned>(pos) != i)
+  ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout);
+  for (size_t i = 0; i < count; i++) {
+    if (pos != -1 && static_cast<size_t>(pos) != i)
       s4u_comms[i]->add_ref();
   }
   return pos;