The Json Serialization currently doesn't support AZStd::any, AZStd::variant and AZStd::optional due to various JSON formatting concerns. This wasn't properly reported resulting in confusion about whether the missing functionality is intentional or a bug. This change makes this explicit with an error message.
As this solution uses a custom json serializer to report the issue, an option was added so it's possible to overwrite a serializer for a specific type.
This doesn't mean that the three listed types will never be supported. If/when a suitable format is found an implementation will be added.
* formatting changes to AzToolsFramework viewport related types + API comment style updates
* minor format change - include ordering
* improve formatting by moving comment
* fix compile error and switch to use AZ_Printf
* small polish changes after review feedback
* Fix recursive attempts to open the log file in the GameLauncher
The AzFramework Application has been updated to default the @user@ and
@log@ aliases to the <engine-root>/user and <engine-root>/user/log
folder respectively if a project isn't set.
Fixed the SystemFile class to support negative offsets if Seek() as per
standard seek function such as fseek
Updated the CrySystem CLog class to use SystemFile instead of FileIOBase
to avoid any asserts that would cause CLog::OpenFile to be recursively
called infinitely
* Removing unused Force Closed variable
* AZ::IO::SystemFile build fixes for Unix platforms. Added a copy constructor for LUAEditorContextInterface.h to fix the LuaEditor build
* Adding missing includes to the WindowsAPI and Android SystemFile headers
* Fixes locating the project dll when using SDK
SDK engine usage has project dll in the project build path, but
searching for module filepaths for loading would have a passing
SystemFile::Exists check but no full filepath was amended to the module.
This causes the module to fail to load.
* Fix locating project module for UnixLike platforms
Fixes the issue with project-centric workflows running GameLauncher, and
it opens AP which can't find the project dynamic module. From AP's
perspective, the project module is not in the executable directory,
which is in engine bin. The SystemFile::Exists check is true on the
file because it uses the 'cwd'. In that situation, an absolute path
must be obtained for the module to be loaded.
* Add missing header to fix UnixLike builds
* Applies a suggested change from PR
Use operator-> on the AZStd::optional
* Add semicolon to a class macro line
Prevent auto formatting indenting the following line.
This set of changes enables conversions for singly-nested slices. Multiple nesting hierarchies are only partially supported at this point. Conversion is also significantly more deterministic, which makes it easier to convert single slices without needing to reconvert every slice or level that relies on it as well.
Changes:
- Added version of Instance::AddInstance() that takes in an alias to allow for deterministic aliases
- Added a "SliceConverterEditorEntityContextComponent" that's used to specifically disable entity activation on creation. The disabling is done this way vs adding a new public API, because the disable shouldn't be required in any normal case outside of this tool.
- Disabled more AWS gems for the SliceConverter, as they're unneeded and cause issues if they're around in the tool.
- Added a small null check to the Camera Controller.
- Added the actual support for slice instance conversion. This instantiates the entities, applies the data patches, turns them into a prefab instance, and generates a JSON patch out of the changes.
* Generic Multi Function Call ability added to extensible nodes
* Code gen improvements, including allowing for more manually codewritten extension of codegen facilities
* CVAR to disable automatic update of deprecated node
* Fixed variable sorting error that can apply to parser/runtime added variables
* Made Edit/SerializeContext ClassBuilder public, as it was needlessly private
* Fixed dangerous Datum::GetValueAddress(), it now checks for an empty storage AZSTd::any, as does Datum::Empty()
The following was changed:
- The remapper in AZ::IdUtils now has an additional argument to tell it what to do when it encounters the same source entity id. The original behavior of ignoring the new entity id and returning the first occurrence is the default. The alternative behavior is to store the last known entity id and return that instead.
- Split the optional arguments for SpawnAllEntities and SpawnEntities.
- SpawnEntities now has an option to continue with the entity mapping from a previous spawn call or to start with a fresh mapping. The latter is the default as the former will come at a performance cost since the mapping table has to be reconstructed.
- Entities spawned using SpawnEntities and ReloadEntities now also get the correct entity mapping applied.
- Added several new unit tests to cover most of the new functionality.
- Fixed some places where the older API version was still called.
A name like Camera.dll was matching against
Atom_Component_DebugCamera.dll so it thought the module was already seen
and wouldn't add it to the list of dynamic modules to load.
The serialize context allows users to reflect fields that are enums to
a class without reflecting the enum type itself with the EnumBuilder. In
this case, the serialize context stores the mapping of the enum's typeid
to the underlying type's typeid. When asking for the class data for the
enum typeid, the underlying type's class data is returned. This was
throwing off the json serializer, which would then see that the type was
"unsigned int" instead of an enum, and attempt to load the unsigned int
value. The unsigned int deserializer would then complain, because the
incoming typeid was the typeid of the enum, and not equal to the typeid
of unsigned int.
This change adds support for detecting the non-reflected enum, and
loading it properly.
Removed unnecessary "is not a function" warnings from ScriptContext.
All these warnings are followed by returning false, which call sites can use to report warnings where appropriate. In the case of material lua functors, it is not appropriate to report a warning which is why I'm removing these. The material system uses the "Call" API to potentially call a function that may or may not exist, and it is acceptable for that function to be absent.
Sometimes deserializing a Json document happens when asset handlers are not
registered. In that case, `FindOrCreateAsset` will fail to create the
asset, since there's no handler registered to create it. When this happens,
`FindOrCreateAsset` returns an Asset instance with a null asset id. This
effectively causes the json deserializer to lose that data, even in
situations where the the actual asset data doesn't need to be loaded, but
the asset id needs to be preserved.
All these warnings are followed by returning false, which call sites can use to report warnings where appropriate. In the case of material lua functors, it is not appropriate to report a warning which is why I'm removing these. The material system uses the "Call" API to potentially call a function that may or may not exist, and it is acceptable for that function to be absent.
While trying to process all slices and levels in Automated Testing, a few bugs came up that needed to be addressed:
- [LYN-3832] TransformComponent had a field removed without updating the version number and converter, which caused a lot of excessive warnings
- SliceComponent would crash in debug builds on instantiation failures due to a null dereference that was guarded against in most but not all places
- SliceConverter now detects when nested slices exist and gracefully warns about it.
- InstanceUpdateExecutor / TemplateInstanceMapper will now immediately remove instances that are unregistered, so that any in the queue don't get processed on a subsequent tick. This was causing crashes when the instance was destroyed before the processing occurred. It also has a side benefit of preventing the same instance from executing multiple times.
- Minor logic bugfix to the pack close warning, the boolean check was flipped.
Also added an early-out on SetTemplateId, since this was causing some unnecessary instance queue entries.
* Updated the SettingsRegistry CommandLineArgumentSettings delimiter function to no longer work on a character basis, but on a line basis.
This provides more control by the user to determine how to the argument into the JSON pointer and JSON value parts
Added function o the Settings Registry to specify the JSON Apply Patch settings to use when performing a JSON Patch/Merge Patch operation
Fixed the SettingsRegistryImpl::CommandLineArgument function to properly set unsigned 64-bit values into the SettingsRegistry by checking the ERRNO of strtoll and strtoull
* Updated the reporting of the JSON Patching operations to supply a JSON pointer of the patched element to the Issue Reporting callback when the patch operation is successful. This allows using the Issue Reporting callback as a notification system when field is updated during a patch operation
* Added support to the AZ Console to be able to run Console Commands based
on notifications from the Settings Registry when a field underneath the
"/Amazon/AzCore/Runtime/ConsoleCommands" object is modified.
This takes advantage of the Settings Registry RegisterNotifier API to
determine when a field is modified as well as the JSON Merger
JsonApplyPatchSettings Issue Reporting Callback to determine when a
field is modified or updated.
As a Side Note also fixed an issue with the AZ Console incorrectly
converting unsigned 64-bit types using strtoll
* Making the Console constructor which accepts an AZ::SettingsRegistryInterface explicit
* Updating string format calls which use *.s for formatting string_views, to use the AZ_STRING_ARG macro
* Addressed typos in comments around the SettingsRegistry AZ Console functions
* Fixed the SettingsRegistryTest that look for an empty value
* clang 6.0.0 constexpr build fix. For some reason clang cannot make a constexpr AZStd::string_view out of a constexpr AZStd::fixed_string despite there being a valid constexpr operator AZStd::string_view
* Mac build fix
* SettingsRegistryTest.MergeSettingsFolder_ConflictingSpecializations_ReportsErrorAndReturnsFalse test fix on Mac
* Updated the LoadSettingsFile test to validate running a console with 0 arguments
Replace the static_cast in the ConsoleTypeHelpers.inl code to convert a str to long long with an aznumeric_cast
* Added printf logging to the ConsoleCommandKeyNotificationHandler to determine if the console commands are being performed on the Jenkins Linux node
* Fixed Dangling string_view reference in the ConsoleCommandKeyNotificationHandler that was causing command execution from a file to fail.
Renamed the second TestFreeFunc function in the ConsoleTests.cpp to validate thath the first TestFreeFunc function is being tested
* Updated the Component Application AZ Console to use the SettingsRegistry as the backend when loading config and Settings Rgistry json files
* Updated the DynamicModuleHandle code to search within the
SettingsRegistry for the FilePathKey_ProjectBuildPath setting in order
to determine the binary directory for the Project.
This is used to locate shared libraries and executables built by the
project when running and Engine SDK binary from within the Engine SDK
* Added a generation step within the Projects.cmake file to generate a
.setreg file containing the CMake build directory root.
The file is output to the <project-root>/user/Registry/build_path.setreg
file.
This occurs only on non-host platforms when configuring for
non-Monolithic builds, since the Build Directory is used to located the
project binary directory in order to located the project's generated
${CMAKE_BINARY_DIR}/bin/$<CONFIG>/Registry directory containing the
cmake_dependencies.<project-name>.<application-name>.setreg file
containing the list of gem modules to load for a given application
Updated the SettingsRegistryMergeUtils AddRuntimeFilePaths function to
be read in the new "/Amazon/Project/Settings/Build/project_build_path"
and use that to form an absolute path to the project build directory by
appending it to the FilePathKey_ProjectPath key.
That key is set in the 'FilePathKey_ProjectBuildPath' constant
Next updated the SettingsRegistryMergeUtils
MergeSettingsToRegistry_TargetBuildDependencyRegistry function to look
within the project build directory to locate the
cmake_dependencies.*.setreg file to load
Tweaked the Settings Registry merge order of the ComponentApplication,
GameApplication and Settings Registry builder to merge the command line
after merging the global user registry and after merging the project
user registry.
Moved the call to MergeSettingsToRegistry_TargetBuildDependencyRegistry
to occur after the above calls to make sure the properly overriden
projects' user registry was merged in order for the correct project
build path to be stored in the SettingsRegistry
* Added a ProjectConfigurationBinPath key which contains the path to the <build-dir>/bin/$<CONFIG> directory for a project which is used to load gem dlls and the Registry/cmake_dependencies.*.setreg files when using an pre-built Editor/AssetProcessor on an external project
* Fixed variable reference to fileNamePath variable
* Removing the default Project Build Path from the Settings Registry
Runtime Filepaths.
Any paths would have to be set explicitly via .setreg/.setregpatch file
or the --project-build-path parameter
Updated the setting of the project build path and project binary
directory to perform existance checks on the paths before setting the
keys of /Amazon/AzCore/Runtime/FilePaths/ProjectBuildPath and
/Amazon/AzCore/Runtime/FilePaths/ProjectConfigurationBinPath
Added a backup project binary path of
<project-build-path>/bin/$<PLATFORM>/$<CONFIG> which is used if the path
of <project-build-path>/bin/$<CONFIG> has not been found
Fixed compile error in DynamicModuleHandle_Apple.cpp
* UnixLike Platform Build fix for the DynamicModuleHandle code
The AssetContainer was getting "stuck" in the case that it tried to load a missing asset that was already registered with the AssetManager as missing. This fixes the bug, as well as adding a unit test for the specific condition.
The AssetContainer was getting "stuck" in the case that it tried to load a missing asset that was already registered with the AssetManager as missing. This fixes the bug, as well as adding a unit test for the specific condition.
Remove:
- Code/CryEngine/CryCommon/Platform
- Some unused Code/CryEngine/CryCommon/Mock files
- Code/Tools/CryXML and almost all of Code/Tools/CryCommonTools
- Code/Tools/TestBed/ResourceCompilerImage
- Tools/DeepBandwidthToExcel
- Various .p4ignore files
* use new ViewportContext interface to set camera transform on load
* WIP fixes for camera viewport handler callbacks
* disable synchonization with old camera when new camera system is enabled
* further updates to camera-input
* ensure event is signalled when camera transform is set
* updates to ModernViewportCameraController
* fix for right click menu appearing with camera
* updates following review feedback
* convert std:: usage to AZStd::
* fix project centric set current
* Adding newline to project.json template
* Adding newline to engine.json
* Changed register -this-engine to be data driven
* adding build server support
* fix log dir
* fix add remove
* fix template creation for restricted
* fix typos and descriptions
* Add newline to the end of template.json
* Adding newline to the end of assets_scan_folders.seteg
* current_project in global project that creates/edits the .o3de/Registry/bootstrap.setreg
* fix the build server flags
* fix the o3de manifest server portion
* disable project manager tests for now. Its changed too much and lytestools is not working at the moment. I will get back to this later.
* fix the Mac build config
* disable project_test, this is the project manager. It has changed too much for these tests to work at the moment. I'll get back to them
Co-authored-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* add overload to ActionManager to support capturing an AZStd::function
* move snapping settings to new settings registry
* remove unneeded reference in ViewportSettings
* move viewport setting function implementations to .cpp file
* add more sensible default values for snapping
* fix variable name for angle snapping
* remove const from function prototype value parameters
* add import/export api for free functions
* change from std::bind to a lambda
* remove redundant const for constexpr string_view
* add AZStd alias for std::abs
* {LYN-2257} Helios - Add Matrix math type to the JSON Serialization system
* Helios/Systems - Add Matrix math type to the JSON Serialization system
* supports both YPR+scale+translation and array of values
* supports Matrix3x3, Matrix3x4, Matrix4x4
Jira: https://jira.agscollab.com/browse/LYN-2257
Tests: Added Serialization/Json/MathMatrixSerializerTests.cpp
* clang compile fixes
* removed typename
* fixing both tests and some default conformity impls
* stablized the comformity test values; rotations drift too much
- Add operators +, -, * and / too all matrix classes
- Add RetrieveScaleSq and GetReciprocalScaled to all matrix classes
- Add unit tests to all matrix classes
- Fix a bug that causes release configuration not to compile.
The behavior of smart pointers was not updated after the changes to raw pointers, in particular with how default JSON Objects were handled. They both now treat an empty JSON Object (a.k.a. the id for default instances) as meaning to create a default instance on the pointer even if the default is a nullptr. To have a nullptr a JSON Null has to be explicitly written.
By setting the initialization flag before the CheckReady() call, the notifications get sent correctly, and no extra asset references remain. This checkin also includes a unit test for the AssetManager that specifically forces this condition to happen and validates that it works correctly.
- Added operator+(Matrix3x4), operator*(float), RetrieveScaleSq and GetReciprocalScaled to Matrix3x4. Used by Cloth CPU Linear Skinning. These operation will be performant as they use SIMD.
- Modified so there are no virtual functions calls at vertex level.
- Caching indices to simplify the loop when applying skinning.
- Caching static variables Matrix3x4 zero and DualQuaternion zero to avoid creating it for every vertex.
- Removing branching to skip joints when the weight is zero, these cases are rarely and this improves performance by removing branching from loops at vertex level.
- Changing skinning influences so it's a continuous block of memory.
- Caching the vector size() if a variable instead of directly using it in a for loop.
The specific mutex is "m_assetMutex" in the AssetManager. AssetManager::ReloadAssetFromData was holding this lock while calling AssignAssetData, which goes on to call OnAssetReloaded. This is dangerous, because any arbitrary logic can be executed during that call, including blocking asset loads, which is what happens in this specific deadlock case. The fix is to move the call to AssignAssetData outside of the mutex lock. AssignAssetData itself already grabs the assetMutex to protect the parts of its logic that are necessary, and then specifically releases the lock before calling OnAssetReloaded, so this change shouldn't introduce any dangerous threading conditions.
(cherry picked from commit 1581a28339f12d521955edef890acfb678a0df6f)
* LYN-2537 Moved the Engine and Editor folder to be within the EngineAssets folder
* Fixed Documentation in bootstrap.cfg to correct the path to the user project specific registry file
* Adding a newline to the output of AssetCatalog 'Registering asset..., but type is not set' message
* Updating the AssetProcessorPlatformConfig.setreg Scan Folder to detect
the @ENGINEROOT@/EngineAssets/Engine path for engine runtime assets and
@ENGINEROOT@/EngineAssets/Editor path for engine tool assets
* Updating references to Icons and other assets to account for moving the
Engine and Editor folder under a single EngineAssets folder
* Moving the Engine Settings Registry folder from Engine/Registry -> Registry
* Removed the LY_PROJECT_CMAKE_PATH define as it is not portable to other locations. It is hard coded to the project location that was used for the CMake configuration. Furthermore it paths with backslashes within it are treated as escape characters and not a path separator
* Updated the LyTestTools asset_processor.py script to copy the exclude.filetag from the EngineAssets/Engine directory now
* Fixed Atom Shader Preprocessing when running using an External Project
* Updated the TSGenerateAction.cpp to fix the build error with using a renamed variable
* Updated the Install_Common.cmake ly_setup_others function to install the
EngineAssets directory and the each of the Gem's Assets directory while
maintaining the relative directory structure to the Engine Root
Also updated the install step to install the Registry folder at the
engine root
* Fixed the copying of the Registry folder to be in the install root, instead of under a second 'Registry' folder
* Moving the AssetProcessorPlatformConfig.setreg file over to the Registry folder
* Updated the LyTestTools and C++ code to point that the new location of
the AssetProcessorPlatformConfig.setreg file inside of the Registry
folder
* Renamed Test AssetProcessor*Config.ini files to have the .setreg extension
* Converted the AssetProcessor test setreg files from ini format to json
format using the SerializeContextTools convert-ini command
* Updated the AssetProcessor CMakeLists.txt to copy over the test setreg files to the build folder
* Updated the assetprocessor test file list to point at the renamed AsssetProcessor*Config setreg filenames
* Removed the Output Prefix code from the AssetProcessor. The complexity that it brought to the AP code is not needed, as users can replicate the behavior by just moving there assets underneath a another folder, underneath the scan folder
* Adding back support to read the AssetProcessorPlatformConfig.setreg file from the asset root. This is only needed for C++ UnitTests as they run in an environment where the accessing the Engine Settings Registry is not available
* Updating the Install_common.cmake logic to copy any "Assets" folder to
the install layout.
The Script has also been updated to copy over the "Assets" folder in the
Engine Root to the install layout instead of an "EngineAssets" folder
* Updating References to EngineAssets source asset folder in code to be the Assets source folder
* Moved the Engine Source Asset folder of 'EngineAssets' to a new folder name of 'Assets'. This is inline with the naming scheme we use for Gem asset folders
* Adding the EngineFinder.cmake to the AutomatedTesting project to allow it to work in a project centric manner
* Updating the LyTestTools copy_assets_to_project function to be able to copy assets with folders to the temporary project root
Fixed an issue in LyTestTools where the temporary log directory could have shutil.rmtree being called twice on it leading to an exception which fails an automated test
Updated the asset_procesor_gui_tests_2 AddScanFolder test to not use the
output prefix, but instead place the source asset root into a
subdirectory
* Correct the AssetProcessorPlatformConfig Scan Folders for the EngineAssets directory to point at the Assets directory
* Updated the asset procesor batch dependency test scan folder to point at the 'Assets' folder instead of 'EngineAssets'
* Setup NativeUIRequests as an AZ::Interface. Adds a NativeUISystemComponent to AzFramework Application.
* Renames NativeUISystemComponent (class) to NativeUISystem, since it's no longer a Component.
* Minor update to SettingsRegistryInterface::Remove doc-comments for accuracy.
* Fixes to make an early fatal shutdown of Editor occur without crash.
* LYN-2723: Updates startup to handle errors: engine root is empty, no valid project.json found (mismatched engine name), or bad project path (launch project picker dialog).
* LYN-2723: Minor formatting/spelling edits.
* LYN-2723: Moves ParseCommandLine from ComponentApplication to SettingsRegistryMergeUtils so it can be used in more places.
* Misc fixes. 'wait_for_connect' setting wasn't being properly applied to AP connection settings. Fix infinite loop in CCmdLine::Next.
* LYN-2723: Addresses review feedback.
* LYN-2723: Reverts some changes that caused a unit test to fail.
* LYN-2723: Reverts one more change that was unnecessary.
The code that adds a $type field for pointers where needed was still assuming that custom serializer were always for primitives, which isn't the case anymore. This changes updates the behavior to allow $type to be added to those as well as long as they use an object. This does now however rely more heavily on earlier checks that the data needs a $type because it otherwise can't tell the difference between a primitive getting a default value (an empty object). In the original code this situation would have resulted in failed serialization though, so it's unlikely to be a problem.
* Asset<T>::QueueLoad didn't trigger any loads in the case where an asset was in a Queued state, it simply returned the reference. This caused problems in the case where an asset was in the process of being cancelled and garbage collected, as it could be in a queued state with nothing actively loading it. The method now detects this case and calls GetAsset(), which triggers a new load.
* AssetContainer::IsValid() was returning true for canceled containers that no longer had a root asset. Now it returns false, to help ensure the container doesn't try to get reused.
* AssetContainer would add entries to the preloadList even if any potential preloads were filtered out from the load. They are no longer added, since they shouldn't be waiting for any dependent assets to load. (This could cause incorrect warnings to print in some situations)
* AssetContainer was erroneously warning about removing assets from a missing waiting list. The warning was removed, as the condition could occur when the same asset was being loading by two different containers - once with dependencies and once without.
* AssetDataStream::RequestCancel has been added, as it was missing, but nothing currently needs to use it.
* AssetManager::GetAssetContainer() now verifies that the container is valid before attempting to reuse it. This prevents asset containers that are in the middle of cancellation from getting reused.
the project root and engine root to fix issues with running the Editor
or AssetProcessor from within the project folder overriding the
project_path with the engine root bootstrap.cfg project_path entry
The order in which the project path is overridden as follows
1. The <engine-root>/bootstrap.cfg is first merged into the Settings
Registry. Any '/Amazon/AzCore/Bootstrap/project_path' would be used
if the following steps don't override that key.
2. Followed by general *.setreg/*.setregpatch files being merged into
the Settings Registry which can override the
'/Amazon/AzCore/Bootstrap/project_path' key
3. Next a project.json file searched upwards from the current executable
directory to determine the project path
4. Finally if a command line parameter that overrides the project path
is supplied it is used instead
I wasn't able to reproduce the deadlock, but from the reported callstack, the following lock inversion happens:
* EditorSurfaceDataSystemComponent::OnCatalogLoaded locked the AssetCatalogRequestBus mutex by calling EnumerateAssets, and then locked m_assetMutex inside GetAsset->FindOrCreateAsset inside the enumerate callback.
* Loading threads would lock m_assetMutex in AssetManager::ValidateAndRegisterAssetLoading, then lock the AssetCatalogRequestBus inside the Asset<T> copy constructor when calling UpdateDebugStatus when the constructor calls SetData->UpgradeAssetInfo->UpdateAssetInfo->AssetCatalogRequestBus::GetAssetInfoById
This should solve the lock inversion on both sides of the problem:
* UpdateDebugStatus now takes in a const ref instead of a copy, so the copy constructor isn't called.
* EditorSurfaceDataSystemComponent::OnCatalogLoaded is rewritten to call GetAsset outside of the enumeration call.
As a bonus, this also removes the blocking load call. The rest of the code already supports asynchronous refreshes as the list assets are added / modified / removed, so this code was changed to leverage the asynchronous refreshes as well.
* removed the script rule from the Editor, now will only be supported via a script or JSON manual edits
* Mesh Serialization - scriptProcessorRule doesn't save with field empty, but produces no error
* added a test to make sure Script Processor Rule operates with an empty filename
Jira: https://jira.agscollab.com/browse/ATOM-14889
Tests: Launched the Editor to removed the script rule from the Editor
AMZN-koppersr
Esteban Papp
srikappa-amzn
srikappa
chcurran
Tom "spot" Callaway
Tom Hulton-Harrop
lumberyard-employee-dm
carlitosan
Eric Phister
Mike Balfour
greerdv
sconel
greerdv
sconel
pappeste
phistere
Terry Michaels
pruiksma
SJ
Chris Burel
Chris Santora
Chris Santora
AMZN-puvvadar
puvvadar
bosnichd
Aaron Ruiz Mora
Nicholas Van Sickle
mbalfour
sharmajs-amzn
Colin Byrne
karlberg
jackalbe
jonbeer
luissemp
hultonha
mnaumov
amzn-mike
amzn-mike
AMZN-AlexOteiza
alexpete
kberg-amzn
jackalbe