Kokopelli crashes on svg export

[James-Fletcher]

The file below reliably causes kokopelli to crash when you try to export to sag. I've downloaded and built the most recent sources on my Mac (Mac OS 10.9.2) and I get the following error:

python(81590,0x11bcc9000) malloc: * error for object 0x7fe1aac86ef0: pointer being freed was not allocated * set a breakpoint in malloc_error_break to debug Abort trap: 6

This is a reduced version of the file narrowing down the error. Removing the final 'slot' (slotl2) in the file below or adjusting the width of the 'slot' makes the export work.

============ .ko file ==============

# Build all parts centered on origin unless otherwise stated

from koko.lib.shapes import *
from math import sqrt

# One cad unit is 1mm
cad.mm_per_unit = 1

# Sizes for 12mm ply on shopbot with 6mm downcut tool
material_thickness = 12 # 12mm ply, measures as 11.6mm to 12mm
initial_material_thickness = material_thickness
tool_size = 6

scale = 50 #one castle unit is 5cm

wall_height = 2*scale
wall_width = 7*scale

tab_height = wall_height/2 # tab should be 10mm high
initial_tab_height = tab_height
tab_width = material_thickness #scale/2

wall = rectangle(-(wall_width/2), wall_width/2, -wall_height/2, wall_height/2)

tab = rectangle(0,tab_width, -(tab_height/2),tab_height/2)
dog = circle(0,0, tool_size/2)
dog_offset = tool_size/3

# tab sticking out of left side
wall += move(tab, -((wall_width/2) + (tab_width)), 0)
wall -= move(dog, -((wall_width/2)+dog_offset), -((tab_height/2)+dog_offset))
wall -= move(dog, -((wall_width/2)+dog_offset), ((tab_height/2)+dog_offset))

# tab inset into right side
#wall -= move(tab, (wall_width/2)-tab_width, 0)
#wall -= move(dog, ((wall_width/2)-tab_width+dog_offset), -((tab_height/2)-dog_offset))
#wall -= move(dog, ((wall_width/2)-tab_width+dog_offset), ((tab_height/2)-dog_offset))

#
# reduce slot width
#

# l2 - exact - 0.4mm width
#material_thickness = initial_material_thickness - 0.4
material_thickness = initial_material_thickness - 0.4
slotl2 = rectangle(-(material_thickness/2), material_thickness/2, -(tab_height/2),tab_height/2)
slotl2 += move(dog, -((material_thickness/2)-dog_offset), -((tab_height/2)-dog_offset))
slotl2 += move(dog, -((material_thickness/2)-dog_offset), ((tab_height/2)-dog_offset))
slotl2 += move(dog, ((material_thickness/2)-dog_offset), -((tab_height/2)-dog_offset))
slotl2 += move(dog, ((material_thickness/2)-dog_offset), ((tab_height/2)-dog_offset))
wall -= move(slotl2, -(initial_material_thickness*6), 0)

cad.shape = wall

[mkeeter]

I've reproduced this bug and am working on tracking it down.

[mkeeter]

Alright, this is a messy bug related to how we contour objects. It looks like your model's expression has zeros on the borders of the spatial partitioning scheme, which breaks multi-scale contouring in subtle ways.

Unfortunately, I haven't touched this code since May of last year, and the code is full of comments like

// Upgrade disconnected edges from children
// (necessary for multi-scale path merging; trust me on this)

This requires a pretty involved fix that I'm unlikely to get to anytime soon. If you're interested in fixing it yourself, take a look at contour.c. I can give you a more detailed explanation of the failure but no easy solution.

Otherwise, try exporting your model at a slightly different scale (i.e. 10.01) or tweaking the parameters by a fraction of a percent; anything that gets the equation's zero points off of the boundaries will solve the crashing.

[James-Fletcher]

Ok, thanks Mat. I may have a dig around myself. If there’s nothing obvious I’ll probably just tinker fractionally with the scale.

James Fletcher

On 27 Mar 2014, at 01:15, Matt Keeter notifications@github.com wrote:

Alright, this is a messy bug related to how we contour objects. It looks like your model's expression has zeros on the borders of the spatial partitioning scheme, which breaks multi-scale contouring in subtle ways.

Unfortunately, I haven't touched this code since May of last year, and the code is full of comments like

// Upgrade disconnected edges from children // (necessary for multi-scale path merging; trust me on this) This requires a pretty involved fix that I'm unlikely to get to anytime soon. If you're interested in fixing it yourself, take a look at contour.c. I can give you a more detailed explanation of the failure but no easy solution.

Otherwise, try exporting your model at a slightly different scale (i.e. 10.01) or tweaking the parameters by a fraction of a percent; anything that gets the equations zero points off of the boundaries will solve the crashing.

— Reply to this email directly or view it on GitHub.