#!/usr/bin/env python
"""
TODO: This only snaps selected elements, and if those elements are part of a
group or layer that has it's own transform, that won't be taken into
account, unless you snap the group or layer as a whole. This can account
for unexpected results in some cases (eg where you've got a non-integer
translation on the layer you're working in, the elements in that layer
won't snap properly). The workaround for now is to snap the whole
group/layer, or remove the transform on the group/layer.
I could fix it in the code by traversing the parent elements up to the
document root & calculating the cumulative parent_transform. This could
be done at the top of the pixel_snap method if parent_transform==None,
or before calling it for the first time.
TODO: Transforming points isn't quite perfect, to say the least. In particular,
when translating a point bezier curve, we translate the handles by the same amount.
BUT, some handles that are attached to a particular point are conceptually
handles of the prev/next node.
Best way to fix it would be to keep a list of the fractional_offsets[] of
each point, without transforming anything. Then go thru each point and
transform the appropriate handle according to the relevant fraction_offset
in the list.
i.e. calculate first, then modify.
In fact, that might be a simpler algorithm anyway -- it avoids having
to keep track of all the first_xy/next_xy guff.
TODO: make elem_offset return [x_offset, y_offset] so we can handle non-symetric scaling
------------
Note: This doesn't work very well on paths which have both straight segments
and curved segments.
The biggest three problems are:
a) we don't take handles into account (segments where the nodes are
aligned are always treated as straight segments, even where the
handles make it curve)
b) when we snap a straight segment right before/after a curve, it
doesn't make any attempt to keep the transition from the straight
segment to the curve smooth.
c) no attempt is made to keep equal widths equal. (or nearly-equal
widths nearly-equal). For example, font strokes.
I guess that amounts to the problyem that font hinting solves for fonts.
I wonder if I could find an automatic font-hinting algorithm and munge
it to my purposes?
Some good autohinting concepts that may help:
http://freetype.sourceforge.net/autohinting/archive/10Mar2000/hinter.html
Note: Paths that have curves & arcs on some sides of the bounding box won't
be snapped correctly on that side of the bounding box, and nor will they
be translated/resized correctly before the path is modified. Doesn't affect
most applications of this extension, but it highlights the fact that we
take a geometrically simplistic approach to inspecting & modifying the path.
"""
from __future__ import division
import sys
# *** numpy causes issue #4 on Mac OS 10.6.2. I use it for
# matrix inverse -- my linear algebra's a bit rusty, but I could implement my
# own matrix inverse function if necessary, I guess.
from numpy import matrix
import simplestyle, simpletransform, simplepath
# INKEX MODULE
# If you get the "No module named inkex" error, uncomment the relevant line
# below by removing the '#' at the start of the line.
#
#sys.path += ['/usr/share/inkscape/extensions'] # If you're using a standard Linux installation
#sys.path += ['/usr/local/share/inkscape/extensions'] # If you're using a custom Linux installation
#sys.path += ['C:\\Program Files\\Inkscape\\share\\extensions'] # If you're using a standard Windows installation
try:
import inkex
except ImportError:
raise ImportError("No module named inkex.\nPlease edit the file %s and see the section titled 'INKEX MODULE'" % __file__)
Precision = 5 # number of digits of precision for comparing float numbers
MaxGradient = 1/200 # lines that are almost-but-not-quite straight will be snapped, too.
class TransformError(Exception): pass
def elemtype(elem, matches):
if not isinstance(matches, (list, tuple)): matches = [matches]
for m in matches:
if elem.tag == inkex.addNS(m, 'svg'): return True
return False
def invert_transform(transform):
transform = transform[:] # duplicate list to avoid modifying it
transform += [[0, 0, 1]]
inverse = matrix(transform).I.tolist()
inverse.pop()
return inverse
def transform_point(transform, pt, inverse=False):
""" Better than simpletransform.applyTransformToPoint,
a) coz it's a simpler name
b) coz it returns the new xy, rather than modifying the input
"""
if inverse:
transform = invert_transform(transform)
x = transform[0][0]*pt[0] + transform[0][1]*pt[1] + transform[0][2]
y = transform[1][0]*pt[0] + transform[1][1]*pt[1] + transform[1][2]
return x,y
def transform_dimensions(transform, width=None, height=None, inverse=False):
""" Dimensions don't get translated. I'm not sure how much diff rotate/skew
makes in this context, but we currently ignore anything besides scale.
"""
if inverse: transform = invert_transform(transform)
if width is not None: width *= transform[0][0]
if height is not None: height *= transform[1][1]
if width is not None and height is not None: return width, height
if width is not None: return width
if height is not None: return height
def vertical(pt1, pt2):
hlen = abs(pt1[0] - pt2[0])
vlen = abs(pt1[1] - pt2[1])
if vlen==0 and hlen==0:
return True
elif vlen==0:
return False
return (hlen / vlen) < MaxGradient
def horizontal(pt1, pt2):
hlen = round(abs(pt1[0] - pt2[0]), Precision)
vlen = round(abs(pt1[1] - pt2[1]), Precision)
if hlen==0 and vlen==0:
return True
elif hlen==0:
return False
return (vlen / hlen) < MaxGradient
class PixelSnapEffect(inkex.Effect):
def elem_offset(self, elem, parent_transform=None):
""" Returns a value which is the amount the
bounding-box is offset due to the stroke-width.
Transform is taken into account.
"""
stroke_width = self.stroke_width(elem)
if stroke_width == 0: return 0 # if there's no stroke, no need to worry about the transform
transform = self.transform(elem, parent_transform=parent_transform)
if abs(abs(transform[0][0]) - abs(transform[1][1])) > (10**-Precision):
raise TransformError("Selection contains non-symetric scaling") # *** wouldn't be hard to get around this by calculating vertical_offset & horizontal_offset separately, maybe 2 functions, or maybe returning a tuple
stroke_width = transform_dimensions(transform, width=stroke_width)
return (stroke_width/2)
def stroke_width(self, elem, setval=None):
""" Return stroke-width in pixels, untransformed
"""
style = simplestyle.parseStyle(elem.attrib.get('style', ''))
stroke = style.get('stroke', None)
if stroke == 'none': stroke = None
stroke_width = 0
if stroke and setval is None:
stroke_width = self.unittouu(style.get('stroke-width', '').strip())
if setval:
style['stroke-width'] = str(setval)
elem.attrib['style'] = simplestyle.formatStyle(style)
else:
return stroke_width
def snap_stroke(self, elem, parent_transform=None):
transform = self.transform(elem, parent_transform=parent_transform)
stroke_width = self.stroke_width(elem)
if (stroke_width == 0): return # no point raising a TransformError if there's no stroke to snap
if abs(abs(transform[0][0]) - abs(transform[1][1])) > (10**-Precision):
raise TransformError("Selection contains non-symetric scaling, can't snap stroke width")
if stroke_width:
stroke_width = transform_dimensions(transform, width=stroke_width)
stroke_width = round(stroke_width)
stroke_width = transform_dimensions(transform, width=stroke_width, inverse=True)
self.stroke_width(elem, stroke_width)
def transform(self, elem, setval=None, parent_transform=None):
""" Gets this element's transform. Use setval=matrix to
set this element's transform.
You can only specify parent_transform when getting.
"""
transform = elem.attrib.get('transform', '').strip()
if transform:
transform = simpletransform.parseTransform(transform)
else:
transform = [[1,0,0], [0,1,0], [0,0,1]]
if parent_transform:
transform = simpletransform.composeTransform(parent_transform, transform)
if setval:
elem.attrib['transform'] = simpletransform.formatTransform(setval)
else:
return transform
def snap_transform(self, elem):
# Only snaps the x/y translation of the transform, nothing else.
# Scale transforms are handled only in snap_rect()
# Doesn't take any parent_transform into account -- assumes
# that the parent's transform has already been snapped.
transform = self.transform(elem)
if transform[0][1] or transform[1][0]: return # if we've got any skew/rotation, get outta here
transform[0][2] = round(transform[0][2])
transform[1][2] = round(transform[1][2])
self.transform(elem, transform)
def transform_path_node(self, transform, path, i):
""" Modifies a segment so that every point is transformed, including handles
"""
segtype = path[i][0].lower()
if segtype == 'z': return
elif segtype == 'h':
path[i][1][0] = transform_point(transform, [path[i][1][0], 0])[0]
elif segtype == 'v':
path[i][1][0] = transform_point(transform, [0, path[i][1][0]])[1]
else:
first_coordinate = 0
if (segtype == 'a'): first_coordinate = 5 # for elliptical arcs, skip the radius x/y, rotation, large-arc, and sweep
for j in range(first_coordinate, len(path[i][1]), 2):
x, y = path[i][1][j], path[i][1][j+1]
x, y = transform_point(transform, (x, y))
path[i][1][j] = x
path[i][1][j+1] = y
def pathxy(self, path, i, setval=None):
""" Return the endpoint of the given path segment.
Inspects the segment type to know which elements are the endpoints.
"""
segtype = path[i][0].lower()
x = y = 0
if segtype == 'z': i = 0
if segtype == 'h':
if setval: path[i][1][0] = setval[0]
else: x = path[i][1][0]
elif segtype == 'v':
if setval: path[i][1][0] = setval[1]
else: y = path[i][1][0]
else:
if setval and segtype != 'z':
path[i][1][-2] = setval[0]
path[i][1][-1] = setval[1]
else:
x = path[i][1][-2]
y = path[i][1][-1]
if setval is None: return [x, y]
def path_bounding_box(self, elem, parent_transform=None):
""" Returns [min_x, min_y], [max_x, max_y] of the transformed
element. (It doesn't make any sense to return the untransformed
bounding box, with the intent of transforming it later, because
the min/max points will be completely different points)
The returned bounding box includes stroke-width offset.
This function uses a simplistic algorithm & doesn't take curves
or arcs into account, just node positions.
"""
# If we have a Live Path Effect, modify original-d. If anyone clamours
# for it, we could make an option to ignore paths with Live Path Effects
original_d = '{%s}original-d' % inkex.NSS['inkscape']
path = simplepath.parsePath(elem.attrib.get(original_d, elem.attrib['d']))
transform = self.transform(elem, parent_transform=parent_transform)
offset = self.elem_offset(elem, parent_transform)
min_x = min_y = max_x = max_y = 0
for i in range(len(path)):
x, y = self.pathxy(path, i)
x, y = transform_point(transform, (x, y))
if i == 0:
min_x = max_x = x
min_y = max_y = y
else:
min_x = min(x, min_x)
min_y = min(y, min_y)
max_x = max(x, max_x)
max_y = max(y, max_y)
return (min_x-offset, min_y-offset), (max_x+offset, max_y+offset)
def snap_path_scale(self, elem, parent_transform=None):
# If we have a Live Path Effect, modify original-d. If anyone clamours
# for it, we could make an option to ignore paths with Live Path Effects
original_d = '{%s}original-d' % inkex.NSS['inkscape']
path = simplepath.parsePath(elem.attrib.get(original_d, elem.attrib['d']))
transform = self.transform(elem, parent_transform=parent_transform)
min_xy, max_xy = self.path_bounding_box(elem, parent_transform)
width = max_xy[0] - min_xy[0]
height = max_xy[1] - min_xy[1]
# In case somebody tries to snap a 0-high element,
# or a curve/arc with all nodes in a line, and of course
# because we should always check for divide-by-zero!
if (width==0 or height==0): return
rescale = round(width)/width, round(height)/height
min_xy = transform_point(transform, min_xy, inverse=True)
max_xy = transform_point(transform, max_xy, inverse=True)
for i in range(len(path)):
self.transform_path_node([[1, 0, -min_xy[0]], [0, 1, -min_xy[1]]], path, i) # center transform
self.transform_path_node([[rescale[0], 0, 0],
[0, rescale[1], 0]],
path, i)
self.transform_path_node([[1, 0, +min_xy[0]], [0, 1, +min_xy[1]]], path, i) # uncenter transform
path = simplepath.formatPath(path)
if original_d in elem.attrib: elem.attrib[original_d] = path
else: elem.attrib['d'] = path
def snap_path_pos(self, elem, parent_transform=None):
# If we have a Live Path Effect, modify original-d. If anyone clamours
# for it, we could make an option to ignore paths with Live Path Effects
original_d = '{%s}original-d' % inkex.NSS['inkscape']
path = simplepath.parsePath(elem.attrib.get(original_d, elem.attrib['d']))
transform = self.transform(elem, parent_transform=parent_transform)
min_xy, max_xy = self.path_bounding_box(elem, parent_transform)
fractional_offset = min_xy[0]-round(min_xy[0]), min_xy[1]-round(min_xy[1])-self.document_offset
fractional_offset = transform_dimensions(transform, fractional_offset[0], fractional_offset[1], inverse=True)
for i in range(len(path)):
self.transform_path_node([[1, 0, -fractional_offset[0]],
[0, 1, -fractional_offset[1]]],
path, i)
path = simplepath.formatPath(path)
if original_d in elem.attrib: elem.attrib[original_d] = path
else: elem.attrib['d'] = path
def snap_path(self, elem, parent_transform=None):
# If we have a Live Path Effect, modify original-d. If anyone clamours
# for it, we could make an option to ignore paths with Live Path Effects
original_d = '{%s}original-d' % inkex.NSS['inkscape']
path = simplepath.parsePath(elem.attrib.get(original_d, elem.attrib['d']))
transform = self.transform(elem, parent_transform=parent_transform)
if transform[0][1] or transform[1][0]: # if we've got any skew/rotation, get outta here
raise TransformError("Selection contains transformations with skew/rotation")
offset = self.elem_offset(elem, parent_transform) % 1
prev_xy = self.pathxy(path, -1)
first_xy = self.pathxy(path, 0)
for i in range(len(path)):
segtype = path[i][0].lower()
xy = self.pathxy(path, i)
if segtype == 'z':
xy = first_xy
if (i == len(path)-1) or \
((i == len(path)-2) and path[-1][0].lower() == 'z'):
next_xy = first_xy
else:
next_xy = self.pathxy(path, i+1)
if not (xy and prev_xy and next_xy):
prev_xy = xy
continue
xy_untransformed = tuple(xy)
xy = list(transform_point(transform, xy))
prev_xy = transform_point(transform, prev_xy)
next_xy = transform_point(transform, next_xy)
on_vertical = on_horizontal = False
if horizontal(xy, prev_xy):
if len(path) > 2 or i==0: # on 2-point paths, first.next==first.prev==last and last.next==last.prev==first
xy[1] = prev_xy[1] # make the almost-equal values equal, so they round in the same direction
on_horizontal = True
if horizontal(xy, next_xy):
on_horizontal = True
if vertical(xy, prev_xy): # as above
if len(path) > 2 or i==0:
xy[0] = prev_xy[0]
on_vertical = True
if vertical(xy, next_xy):
on_vertical = True
prev_xy = tuple(xy_untransformed)
fractional_offset = [0,0]
if on_vertical:
fractional_offset[0] = xy[0] - (round(xy[0]-offset) + offset)
if on_horizontal:
fractional_offset[1] = xy[1] - (round(xy[1]-offset) + offset) - self.document_offset
fractional_offset = transform_dimensions(transform, fractional_offset[0], fractional_offset[1], inverse=True)
self.transform_path_node([[1, 0, -fractional_offset[0]],
[0, 1, -fractional_offset[1]]],
path, i)
path = simplepath.formatPath(path)
if original_d in elem.attrib: elem.attrib[original_d] = path
else: elem.attrib['d'] = path
def snap_rect(self, elem, parent_transform=None):
transform = self.transform(elem, parent_transform=parent_transform)
if transform[0][1] or transform[1][0]: # if we've got any skew/rotation, get outta here
raise TransformError("Selection contains transformations with skew/rotation")
offset = self.elem_offset(elem, parent_transform) % 1
width = self.unittouu(elem.attrib['width'])
height = self.unittouu(elem.attrib['height'])
x = self.unittouu(elem.attrib['x'])
y = self.unittouu(elem.attrib['y'])
width, height = transform_dimensions(transform, width, height)
x, y = transform_point(transform, [x, y])
# Snap to the nearest pixel
height = round(height)
width = round(width)
x = round(x - offset) + offset # If there's a stroke of non-even width, it's shifted by half a pixel
y = round(y - offset) + offset
width, height = transform_dimensions(transform, width, height, inverse=True)
x, y = transform_point(transform, [x, y], inverse=True)
y += self.document_offset/transform[1][1]
# Position the elem at the newly calculate values
elem.attrib['width'] = str(width)
elem.attrib['height'] = str(height)
elem.attrib['x'] = str(x)
elem.attrib['y'] = str(y)
def snap_image(self, elem, parent_transform=None):
self.snap_rect(elem, parent_transform)
def pixel_snap(self, elem, parent_transform=None):
if elemtype(elem, 'g'):
self.snap_transform(elem)
transform = self.transform(elem, parent_transform=parent_transform)
for e in elem:
try:
self.pixel_snap(e, transform)
except TransformError, e:
print >>sys.stderr, e
return
if not elemtype(elem, ('path', 'rect', 'image')):
return
self.snap_transform(elem)
try:
self.snap_stroke(elem, parent_transform)
except TransformError, e:
print >>sys.stderr, e
if elemtype(elem, 'path'):
self.snap_path_scale(elem, parent_transform)
self.snap_path_pos(elem, parent_transform)
self.snap_path(elem, parent_transform) # would be quite useful to make this an option, as scale/pos alone doesn't mess with the path itself, and works well for sans-serif text
elif elemtype(elem, 'rect'): self.snap_rect(elem, parent_transform)
elif elemtype(elem, 'image'): self.snap_image(elem, parent_transform)
def effect(self):
svg = self.document.getroot()
self.document_offset = self.unittouu(svg.attrib['height']) % 1 # although SVG units are absolute, the elements are positioned relative to the top of the page, rather than zero
for id, elem in self.selected.iteritems():
try:
self.pixel_snap(elem)
except TransformError, e:
print >>sys.stderr, e
if __name__ == '__main__':
effect = PixelSnapEffect()
effect.affect()
