Sheets carry a managed buffer of data that allows updates to be made without having to constantly fetch and set data to the texture memory of the video card. This is useful for things like SheetBuilder which make small progressive changes to sheets.
However these buffers are often large and are kept alive because sheets are referenced by the sprites that use them. If this buffer is explicitly null'ed when it is no longer needed then the GC can reclaim it. Sometimes a buffer need not even be created because the object using the sheet only works on the texture directly anyway.
In practise, this reduced memory consumed by such buffers from ~165 MiB to ~112 MiB (at the start of a new RA skirmish mission).
- Add separate ImmutablePalette and MutablePalette classes since the distinction is extremely important to HardwarePalette.
- Keep a cache of palettes in HardwarePalette to avoid reallocation them every time ApplyModifiers is called.
- Palettes that are not allowed to be modified are copied to the buffer once when added, rather than every time ApplyModifiers is called.
- The AdjustPalette method now takes a read-only dictionary to prevent the dictionary being messed with.
- Added a constant for the palette size to remove its usage as a magic number in several areas.
- The ColorPreviewManagerWidget is annoying in that it needs to actually permanently update a palette after it has been added. To allow this, HardwarePalette now allows a palette to be replaced after initialization. The WorldRenderer therefore now also updates the PaletteReference it created earlier with the new palette to prevent stale data being used elsewhere.
- Updated implementations to return a ReadOnlyList around an array (to reduce wasted memory from exposing lists or lazy enumerators around lists).
- Protect non-public ISpriteFrame classes by making them inner classes to prevent casting.
- Added an AsReadOnly extension method for lists.
The lookup accounts for ~50-60% of the time spent in GetTerrainIndex and GetTerrainInfo, and these methods themselves can account for up to 1.3% of total CPU used so this is a small but measurable win.
