The analyzer was built to go off syntax nodes. This (AFAICT) meant that the SemanticModel had to be recalculated for every single invocation.
If you don't know what the above means: it basically means the compiler has to re-analyze the entire file.
Fix this by moving it to an operation analyzer so the compiler can properly cache the semantic model.
This was an undocumented breaking change introduced by https://github.com/space-wizards/RobustToolbox/pull/5915. The behavior does not make much sense: you're specifying coordinates relative to the map, so the rotation should be relative to the map too.
Apparently cheat clients have figured out that none of SS14's code does validation against NaN inputs. Uh oh.
IRobustSerializer can now be configured to remove NaN values when reading. This is intended to be set on the server to completely block the issue.
Added "Unsafe" float types that can be used to bypass the new configurable behavior, in case somebody *really* needs NaNs.
An alternative option was to make a "SafeFloat" type, and only apply the sanitization to that. The problem is that would require updating hundreds if not thousands of messages in SS14, and probably significantly confuse contributors on "when use what." Blocking NaNs by default is likely to cause little issues while ensuring the entire exploit is guaranteed impossible.
* WebSocket-based data transfer system
* Move resource downloads/uploads to the new transfer system
Should drastically increase the permitted practical size
* Transfer impl for Lidgren
* Async impl for receive stream
* Use unbounded channel for Lidgren
* Add metrics
* More comments
* Add serverside stream limit to avoid being a DoS vector
* Fix tests
* Oops forgot to actually implement sequence channels in NetMessage
* Doc comment for NetMessage.SequenceChannel
* Release notes
This means we don't have to JIT a bunch of UIs that you might not open, reducing memory usage and startup overhead.
One (1) UI is always JITed in another thread before prototype UIs are loaded, so as to warm up the JIT machinery. Said type is DropDownDebugConsole which always gets used anyways so there's no harm in it.
In total, these changes save more than a second of startup time for me.
* Added basic loading screen
* Make it look better!
* I forgor xD
* Fix test fails
* Add comment
* Removed unused import
* Only write to file if the number of sections changed
* Servers can now have their own settings
* Minor optionzation and rare colors
* Remove some of the cvars
* debug only loading messages
* Added a few more steps
* Only one section at a time
* nullable section name
* Lock out functions if finished
* Get rid of saving the ccvar
* Cleanup
* Forgot!
* A few tweaks
* Disable vsync
* remove colors
* remove outdated vsync functions
* Silly me xD
* What I get for trying to be clever... ;(
* Better seconds display
* Simplify drawing logic + it looks better
* Type does not need to be partial
* Make interface to expose to content
* Use correct define to gate showing debug info
Should be TOOLS instead of DEBUG
* Use appropriate exception type in BeginLoadingSection
* Fix exception when closing window during loading screen
Would try to stop the main loop before it exists.
* Rename CVars, put debug info behind CVar instead of conditional compilation.
* Add to RELEASE-NOTES.md
* Add UI scaling support
* Make ILoadingScreenManager fully internal
Didn't realize content can't touch it as it'd break the total amount of sections
* Don't re-enable vsync manually, GameController does it at the end of init
* Add command to show top load time usage.
* Improve verbosity of debug time tracking
More steps and some steps named better
---------
Co-authored-by: PJB3005 <pieterjan.briers+git@gmail.com>
* System font API
This is a new API that allows operating system fonts to be loaded by the engine and used by content.
Fonts are provided in a flat list exposing all the relevant metadata. They are loaded from disk with a Load call.
Initial implementation is only for Windows DirectWrite.
* Load system fonts as memory mapped files if possible.
This allows sharing the font file memory with other processes which is always good.
* Use ArrayPool to reduce char array allocations
* Disable verbose logging
* Implement system font support on Linux via Fontconfig
* Implement macOS support
* Add "FREEDESKTOP" define constant
This is basically LINUX || FREEBSD. Though FreeBSD currently gets detected as LINUX too. Oh well.
* Compile out Fontconfig and CoreText system font backends when not on those platforms
* Don't add Fontconfig package dep on Mac/Windows
* Allow disabling system font support via CVar
Cuz why not.
