Simulating a cloned context containing hydroelastic contact results fails

[joemasterjohn]

What happened?

While working on #18057 we came across the following:

• Forward simulate a plant with active hydroelastic contacts to time t.
• Clone the context of that plant at time t.
• Construct a new simulator using the cloned context.
• Forward simulate the new simulator with the cloned context.
• Failure at the next timestep when adding hydroelastic contact results (relevant log below)

Reproduction branch with containing the reference commit sha can be found here: https://github.com/joemasterjohn/drake/tree/hydro_cloned_context_bug

@sammy-tri came up with the proposed fix during review:

diff --git a/multibody/plant/contact_results.cc b/multibody/plant/contact_results.cc
index fd2d3925ba..869665b7d6 100644
--- a/multibody/plant/contact_results.cc
+++ b/multibody/plant/contact_results.cc
@@ -66,11 +66,8 @@ void ContactResults<T>::set_plant(const MultibodyPlant<T>* plant) {
 template <typename T>
 void ContactResults<T>::Clear() {
   point_pairs_info_.clear();
-  if (hydroelastic_contact_vector_ownership_mode() == kAliasedPointers) {
-    hydroelastic_contact_vector_of_pointers().clear();
-  } else {
-    hydroelastic_contact_vector_of_unique_ptrs().clear();
-  }
+  hydroelastic_contact_info_ =
+      std::vector<const HydroelasticContactInfo<T>*>();
   plant_ = nullptr;
 }

Version

6c7163f15334e1cc407dec3c51b7825347566959

What operating system are you using?

Ubuntu 20.04

What installation option are you using?

compiled from source code using Bazel

Relevant log output

abort: Failure at multibody/plant/contact_results.cc:98abort: Failure at multibody/plant/contact_results.cc:98 in AddContactInfo(): condition 'hydroelastic_contact_vector_ownership_mode() == kAliasedPointers' failed. in AddContactInfo(): condition 'hydroelastic_contact_vector_ownership_mode() == kAliasedPointers' failed.

[SeanCurtis-TRI]

Is this possibly related to #17375.