The reported line visibility issue was caused by a wrong calculation of a 2D
bounding box (so-called "proscenium face" in Freestyle) in the case of a
spherical grid data structure used for a perspective camera. The problem was
resulting from the proscenium computation based on two corners (min and max)
of the 3D bounding box of imported mesh data. Aware of the spherical coordinate
transformation involving non-linear (arctangent) functions, now the proscenium
is computed by taking in account all the eight corners of the 3D bounding box.
Also added minor code changes to facilitate future debugging.
The cause of inconsistent edge connectivity in the view map (documented in the commit
log of revision 58006) was identified and fixed. The problem was that when a ViewEdge
was split at a cusp vertex (ViewMapBuilder::computeCusps()), the ViewVertex at one end
of a newly created ViewEdge in ViewMap::InsertViewVertex() was not properly updated to
take account of edge connectivity changes.
Clipping start is negative when the viewport preview is used with the orthographic view,
while Freestyle assumes that imported mesh data are in the camera coordinate system
with the view point located at origin. The present solution is to adjust the clipping
start/end and introduce a Z-axis offset for mesh vertices so as to satisfy the assumption.
Problem report by flokkievids in the BA Freestyle thread, thanks!
Also made changes to suppress warnings in strip creation when Freestyle debugging is disabled.
Now add_freestyle() in pipeline.c takes a second argument to enable/disable
stroke rendering. When stroke rendering is disabled, the function allocates
data structures but does not perform stroke rendering. The allocated data
structures (mostly left unpopulated with data elements) are intended to allow
for the Read Full Sample Layers (Shift-R) command in the compositor.
When the sketchy chaining is used, stroke geometry may contain a 180-degree U-turn.
If the 'error' parameter of the Bezier Curve geometry modifier is small (e.g., 10),
Bezier curve fitting will recursively split the original stroke into two pieces.
This splitting may take place at a U-turn point, causing a numerical singularity issue
that leads to a crash.
Problem report by edna in the BA Freestyle thread, with an example .blend to reproduce
the problem. Thanks a lot!
There were two issues:
- Line visibility computations are very slow in the case of the provided .blend file, which gave
an impression that the rendering process got stuck. The slowness can be explained by the present
data structures used for the line visibility computations, together with the specific mesh distribution
of the test scene. At the moment Freestyle uses a regular grid in the 2D image coordinate system
to divide tris/quads into small groups in order to accelerate the line visibility computations.
On the other hand, the test scene is populated a big plane (made of one quad) and a moderately
detailed mesh object (22K tris). The scale of the latter mesh is animated from nearly zero to
about 0.2 to make the object show up over time. When the scale is nearly equal to zero, all the
tris concentrate in one grid cell, so essentially there is no performance gain from the grid data
structure optimized for speed. It looks like a better grid data structure (possibly based on
adaptive grid refinement) is necessary to genuinely address the identified performance issue. For now
the progress bar of Blender is employed to better inform users of the amount of work done in the line
visibility computations.
- A crash was caused by an excessive memory allocation request. The X and Y dimensions of the grid
data structure are determined based on the average area of mesh faces in the given scene. When the big
plane in the test scene is excluded from the rendering, the average area is almost zero (on the order
of 1e-5). As a result of this extremely small average area, the X and Y dimensions were set to a very
large number, causing a fatal memory allocation error. The present revision has introduced a hard
upper limit to the dimensions of the grid data structure to avoid this kind of numerical instability.
Suitable for inclusion in 2.67a.
Logical AND, OR and XOR operations on freestyle.Nature instances gave an error in some cases.
The updated C implementation of these bitwise operations is intended to reproduce the following Python implementation:
class Nature(int):
def __and__(self, other):
return Nature(int(self) & int(other))
def __or__(self, other):
return Nature(int(self) | int(other))
def __xor__(self, other):
return Nature(int(self) ^ int(other))
The problem report was by plasmasolutions on IRC, thanks a lot!
SpatialNoiseShader, as well as SmoothingShader were not updating stroke length after
geometry modification, causing an infinite loop in Stroke::Resample(int iNPoints) due to
incorrect length-based resampling of stroke vertices.
Reverted the trunk revision 56136 and part of revision 56127 concerning the local Main
for Freestyle and temporary scene generation for stroke rendering.
The function do_merge_fullsample() in pipeline.c has access to the Scene of each Render,
so that the temporary Scene generated by Freestyle has to be kept for FSAA even after
stroke rendering has been done. By the same token, the local Main has been moved from
the BlenderStrokeRenderer class to Render. It is noted that free_all_freestyle_renders()
in pipeline.c is intended to get the temporary Scene of each Render released specifically
taking account of the FSAA case.
It's more like an ownership thing, but it's not so much
clear who actually is an owner of temporary scene, so
for now just ensure Render does not have link to it
after strokes are rendered.
There're also suspicious thing related on why separate
RE_free_database is needed for freestyle renders, but
that's also another story.
This means main database is no longer pollutes with
temporary scene and objects needed for freestyle
render.
Actually, there're few of separated temporary mains
now. Ideally it's better to use single one, but it's
not so much trivial to pass it to all classes. Not
so big deal actually.
Required some changes to blender kernel, to make it
possible to add object to a given main, also to
check on mesh materials for objects in given main.
This is all straightforward changes.
As an additional, solved issue with main database
being infinitely polluted with text blocks created
by create_lineset_handler function.
This fixes:
- #35003: Freestyle crashes if user expands objects in FRS1_Scene
- #35012: ctrl+f12 rendering crashes when using Freestyle
Because of our release soon, feature has been added behind the Debug Menu.
CTRL+ALT+D and set it to -1. Or commandline --debug-value -1.
When debug set to -1, you can put the viewport to 'render' mode, just like
for Cycles. Notes for testers: (and please no bugs in tracker for this :)
- It renders without AA, MBlur, Panorama, Sequence, Composite
- Only active render layer gets rendered. Select another layer will re-render.
- But yes: it works for FreeStyle renders!
- Also does great for local view.
- BI is not well suited for incremental renders on view changes. This only
works for non-raytrace scenes, or zoom in ortho or camera mode, or for
Material changes. In most cases a full re-render is being done.
- ESC works to stop the preview render.
- Borders render as well. (CTRL+B)
- Force a refresh with arrow key left/right. A lot of settings don't trigger
re-render yet.
Tech notes:
- FreeStyle is adding a lot of temp objects/meshes in the Main database. This
caused DepsGraph to trigger changes (and redraws). I've prepended the names
for these temp objects with char number 27 (ESC), and made these names be
ignored for tag update checking.
- Fixed some bugs that were noticable with such excessive re-renders, like
for opening file window, quit during renders.
