This is now replaced by r.SkinCache.Allow (default to true) in case users still want to disable it and claim the memory back
#jira UE-149155
#preflight 627290287b6d0faffa6fc8b4
#rb Wojciech.Krywult Brian.White
[CL 20042354 by benjamin rouveyrol in ue5-main branch]
- Added deletion of empty directories.
- Applied the rate limit to directory scans.
- Reduced the default scan rate by a factor of 10.
#preflight 626c3a8b0f5f22f922511514
#rb Zousar.Shaker
[CL 19984889 by Devin Doucette in ue5-main branch]
#rb Per.Larsson
#rnx
#jira UE-148223
#preflight 6267c4e2853fdb6fddb20be3
- Remove the Core.ContentVirtualizationDebugOptions section from the ini files entirely.
- Remove the ::ApplyDebugSettingsFromConfigFiles method as we no longer load any debug values from the config files.
[CL 19923780 by paul chipchase in ue5-main branch]
This resolves a UBT issue where building from Visual Studio/Rider while the editor is running wouldn't automatically change to HotReload mode.
#preflight 626698887e06ec75059f44b8
#jira none
#rb Ben.Marsh, Jonathan.Adamczewski
#lockdown mark.lintott
#ROBOMERGE-AUTHOR: tim.smith
#ROBOMERGE-SOURCE: CL 19901402 in //UE5/Release-5.0/... via CL 19902726
#ROBOMERGE-BOT: UE5 (Release-Engine-Staging -> Main) (v942-19904256)
[CL 19904470 by tim smith in ue5-main branch]
#rb Per.Larsson
#jira UE-148223
#rnx
#preflight 62666fc90634d0904cca8756
- It was probably a bad idea to have debug settings like this applied from the config file as it would be fairly easy for people to accidently forget about them or submit them.
- The single threaded command can now be set via the commandline on start up using the '-VA-XXX' synctax that the system has started to adopt.
- Additionally it has been exposed as a console command so that it can be toggles on and off at runtime.
- Removed 'ForceSingleThreaded' from BaseEngine.ini and removed the parsing code for this from FVirtualizationManager::ApplyDebugSettingsFromConfigFiles.
- Removed 'FailPayloadPullOperations' from BaseEngine.ini, it hasn't been a valid setting for a while.
- Changed ConsoleCommands to ConsoleObjects and use it to store the base class IConsoleObject instead so it can contain console variables
- When pushing payloads we now only take the singlethreaded lock after checking if we have any backends/are enabled.
[CL 19900622 by paul chipchase in ue5-main branch]
Instead of having two float-related literal nodes, we should only have one. The single-precision variant needs to be deprecated in favor of the double-precision one.
#rb benjamin.fox
#preflight 6255f7ee9741b87f66b6b069
#jira UE-145287
[CL 19757056 by dave jones2 in ue5-main branch]
- Introduced EConflictReportFormat to determine the output format of the conflicts report.
- Updated the localization .ini files to specify the requested output format.
#jira: UE-143696
#rb: Jamie.Dale, Vincent.Gauthier
#preflight: 62573db00b7c7f0f38c6405b
[CL 19746764 by Leon Huang in ue5-main branch]
We neglected to redirect Conv_FloatToText nodes to Conv_DoubleToText, which was causing loss of precision when formatting text. However, this revealed another issue with the format text node.
Since the format text node uses wildcards, it needs a sorting priority of Low_UsesDependentWildcard. Otherwise, linked inputs won't be updated before we sync the node's pins. For example, we might be linked to a Conv_FloatToText node that hasn't yet been converted to a Conv_DoubleToText node. As a result, the format text node will use a single precision float input for the connected conversion node instead of a double.
Another issue is that the format text node is marked as causing a structural Blueprint change. The current sorting function prioritizes that over its refresh priority. Fortunately, we don't need to concern ourselves with structural modification if we're compiling on load, which is when we really need the refresh priority to work as expected.
Finally, K2Node_EnumEquality node needs to have its knot dependencies updated prior to reconstruction. Otherwise, it might fail to find a valid UEnum, and disconnect its inputs. Since knot nodes already have this logic built in, we can just force a type propagation on any linked knots.
#jira UE-141180
#preflight 6255c2ed153828d27337753f
#rb Phillip.Kavan
[CL 19726070 by dave jones2 in ue5-main branch]
- Make FStreamedAudioCacheDerivedDataWorker thread-safe by copying USoundWave members before going async and querying the DDC directly for compressed sound data
- Failure to get streamed chunks from DDC will now try to build asynchronously if allowed to instead of falling back to synchronous behavior by default
- Make LoadZerothChunk launch an async build in editor instead of forcing a wait
- GetNumChunks(), GetChunks() and GetZerothChunkData() will wait on the async build if not finished to return appropriate data to the caller
- Allow DDC fill commandlet to benefit from async build
- Improve cooker logic to reduce waits and avoid preparation of the running platform data during cook if not needed
- Add async sound compiler so that we benefit from its scheduling and ensure sounds are all properly built after a call to FAssetCompilingManager::Get().FinishAllCompilations().
- Protect the streamed audio platform data's AsyncTask property from multi-threaded access
- Can be disabled from editor experimental settings, by ini settings, by cmdline using -asyncsoundwavecompilation=0 or by CVar using Editor.AsyncSoundWaveCompilation=0
- Benchmarks with -DDC-Local-MissTypes=Audio+StreamedAudio on Artemis_Terrain
- 3m52s to 2m03s for map loading
- 8m13s to 2m59s for PIE
#tests
- No binary diff on FN cooked sounds with both recompilation and shared ddc code paths
- All Audio related tests in EngineTest
- As many AudioQA tests as possible
- Lyra -game
#jira UE-142500
#rb Phil.Popp
#preflight 624ee0703661c8f04a3842b8
[CL 19667776 by danny couture in ue5-main branch]
