- Changed PathArcTo to select start/end sample in quatized multiples of arc fast sample count
- This is done in order to avoid small segments at the start and end of the path, which causes issues with line AA ends
- Renamed CalcArcStepAndCount to CalcCornerArcStepAndCount to make it more clear that it returns the values for 90 degree arc
Fixed issue where were creating detailed line textures up to detailed witdh + 1
Changed IM_DRAWLIST_TEX_LINES_DETAILED_WIDTH_MAX to IM_DRAWLIST_TEX_LINES_DETAILED_WIDTH_COUNT which no accounts for all the detailed line textures.
The "does stroke cover the whole shape" used to be overly pessimistic to stroking artifacts. This was due to the fringe, where the geometry would mess up before visually closing the shape.
Fixed the issue by adding max inner offest for the AddPolyline(), which limits the inner offset and also adjusts the texture coords.
Second part of the fix was to calculate the pass the max inner offset from the common shapes (circle, ngon, ellipse, rect).
While this does not fix all the ellipse rendering issues, it limits the issue a lot.
e.g. AddRect() w/ size=5, thickness=2 used to trigger AddRectFilled() fallback because threshold was pessimistic to prevent the AddPolyline() issues.
Require all corners to fit inside the rect to avoid overdraw (or fallback to path rendering).
eg. AddRectFilled() w/ size=10, rounding=8 or 10 would break.
ImDrawFlags_AAEnds and ImDrawFlags_StrokeLegacy ok. (the most important ones).
ImDrawFlags_UseTexForRoundCorners ok.
ImDrawFlags_AAFill, ImDrawFlags_AAStrokes: for primitives: only for selected ones taking flags params. Haven't added the flag to all functions because it feels like a fringe use case.
ImDrawFlags_TextNoPixelSnap supported in RenderText(), per-prim (9417)
- test overestimate before testing more expensive details when testing if stroked ngon is actualy filled
- Note: not a common function, but the test harness is using this to draw a triangle, which makes it show up in profiling
- revert texture selection to use just (inlined) round()
- the values were tuned to hide texture selection pops on small thickness
- since we're using super sampling for the problematic range, the tweaks are not necessayr anymore
- the ImDrawListFlags_AntiAliasedFill and ImDrawListFlags_AntiAliasedLines flags work again
- added non-AA rendering for polyline and convex poly filled
- do not use corner textures when AA is not used
- Add missing no-AA for single line case
- added CalcArcStepAndCount() to calculate the step size consistently
- _AddRectTinyRounding: fixed stem placement with thin lines (clamping was needed for an older version of the function)
- _AddRectTinyRounding: fixed texture coordinates (uses if inner/outer were reversed)
- use polygon instead of circle when calculating baked corner rounds (the circle was more accurate but looked different size comparet to polyline)
- use macro to test if rounding value is worthy
- changed threshold to be a bit smaller (for smoother handling of small radius), and relative to fringe
- Use inline array + AddPolyline/AddConvexPolyFilled for triangle and quad
- do less computation for detecting small shape rendering
- add IM_UNLIKELY for very unlikely branches
- changed truncated tests to use macro, and inline the computation in debug builds (~15% was spent on calling the functions)
- inlined the common rectangle cases in AddRect() for debug build speed up
