#4294 Make upload order more deterministic

to make root prim more deterministic instead of being random each try

1 files changed, 22 insertions, 10 deletions

diff --git a/indra/newview/llmeshrepository.cpp b/indra/newview/llmeshrepository.cpp
index b5a1bbd534..6236df1a13 100644
--- a/indra/newview/llmeshrepository.cpp
+++ b/indra/newview/llmeshrepository.cpp
@@ -2699,10 +2699,21 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
 
     S32 instance_num = 0;
 
-    for (instance_map::iterator iter = mInstance.begin(); iter != mInstance.end(); ++iter)
+    // Upload should happen in deterministic order, so sort instances by model name.
+    // Note: probably can sort by mBaseModel->mSubmodelID here as well to avoid
+    // running over the list twice.
+    std::vector<std::pair<LLModel*, instance_list>> sorted_instances(mInstance.begin(), mInstance.end());
+    std::sort(sorted_instances.begin(), sorted_instances.end(),
+        [](const std::pair<LLModel*, instance_list>& a, const std::pair<LLModel*, instance_list>& b)
+    {
+        return a.first->mLabel < b.first->mLabel;
+    });
+
+    // Handle models, ignore submodels for now
+    for (auto& iter : sorted_instances)
     {
         LLMeshUploadData data;
-        data.mBaseModel = iter->first;
+        data.mBaseModel = iter.first;
 
         if (data.mBaseModel->mSubmodelID)
         {
@@ -2711,7 +2722,7 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
             continue;
         }
 
-        LLModelInstance& first_instance = *(iter->second.begin());
+        LLModelInstance& first_instance = *(iter.second.begin());
         for (S32 i = 0; i < 5; i++)
         {
             data.mModel[i] = first_instance.mLOD[i];
@@ -2758,8 +2769,8 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
         }
 
         // For all instances that use this model
-        for (instance_list::iterator instance_iter = iter->second.begin();
-             instance_iter != iter->second.end();
+        for (instance_list::iterator instance_iter = iter.second.begin();
+             instance_iter != iter.second.end();
              ++instance_iter)
         {
 
@@ -2857,10 +2868,11 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
         }
     }
 
-    for (instance_map::iterator iter = mInstance.begin(); iter != mInstance.end(); ++iter)
+    // Now handle the submodels.
+    for (auto& iter : sorted_instances)
     {
         LLMeshUploadData data;
-        data.mBaseModel = iter->first;
+        data.mBaseModel = iter.first;
 
         if (!data.mBaseModel->mSubmodelID)
         {
@@ -2869,7 +2881,7 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
             continue;
         }
 
-        LLModelInstance& first_instance = *(iter->second.begin());
+        LLModelInstance& first_instance = *(iter.second.begin());
         for (S32 i = 0; i < 5; i++)
         {
             data.mModel[i] = first_instance.mLOD[i];
@@ -2916,8 +2928,8 @@ void LLMeshUploadThread::wholeModelToLLSD(LLSD& dest, bool include_textures)
         }
 
         // For all instances that use this model
-        for (instance_list::iterator instance_iter = iter->second.begin();
-             instance_iter != iter->second.end();
+        for (instance_list::iterator instance_iter = iter.second.begin();
+             instance_iter != iter.second.end();
              ++instance_iter)
         {