inkstitch/lib/elements/element.py

646 wiersze
23 KiB
Python

# Authors: see git history
#
# Copyright (c) 2010 Authors
# Licensed under the GNU GPL version 3.0 or later. See the file LICENSE for details.
import sys
from contextlib import contextmanager
from copy import deepcopy
import inkex
import numpy as np
from inkex import bezier
from ..commands import find_commands
from ..debug import debug
from ..exceptions import InkstitchException, format_uncaught_exception
from ..i18n import _
from ..marker import get_marker_elements_cache_key_data
from ..patterns import apply_patterns, get_patterns_cache_key_data
from ..stitch_plan.lock_stitch import (LOCK_DEFAULTS, AbsoluteLock, CustomLock,
LockStitch, SVGLock)
from ..svg import (PIXELS_PER_MM, apply_transforms, convert_length,
get_node_transform)
from ..svg.tags import INKSCAPE_LABEL, INKSTITCH_ATTRIBS
from ..utils import Point, cache
from ..utils.cache import get_stitch_plan_cache, CacheKeyGenerator
class Param(object):
def __init__(self, name, description, unit=None, values=[], type=None, group=None, inverse=False,
options=[], default=None, tooltip=None, sort_index=0, select_items=None):
self.name = name
self.description = description
self.unit = unit
self.values = values or [""]
self.type = type
self.group = group
self.inverse = inverse
self.options = options
self.default = default
self.tooltip = tooltip
self.sort_index = sort_index
self.select_items = select_items
def __repr__(self):
return "Param(%s)" % vars(self)
# Decorate a member function or property with information about
# the embroidery parameter it corresponds to
def param(*args, **kwargs):
p = Param(*args, **kwargs)
def decorator(func):
func.param = p
return func
return decorator
class EmbroideryElement(object):
def __init__(self, node):
self.node = node
@property
def id(self):
return self.node.get('id')
@classmethod
def get_params(cls):
params = []
for attr in dir(cls):
prop = getattr(cls, attr)
if isinstance(prop, property):
# The 'param' attribute is set by the 'param' decorator defined above.
if hasattr(prop.fget, 'param'):
params.append(prop.fget.param)
return params
@cache
def get_param(self, param, default):
value = self.node.get(INKSTITCH_ATTRIBS[param], "").strip()
return value or default
@cache
def get_boolean_param(self, param, default=None):
value = self.get_param(param, default)
if isinstance(value, bool):
return value
else:
return value and (value.lower() in ('yes', 'y', 'true', 't', '1'))
@cache
def get_float_param(self, param, default=None):
try:
value = float(self.get_param(param, default))
except (TypeError, ValueError):
value = default
if value is None:
return value
if param.endswith('_mm'):
value = value * PIXELS_PER_MM
return value
@cache
def get_int_param(self, param, default=None):
try:
value = int(self.get_param(param, default))
except (TypeError, ValueError):
return default
if param.endswith('_mm'):
value = int(value * PIXELS_PER_MM)
return value
# returns 2 float values as a numpy array
# if a single number is given in the param, it will apply to both returned values.
# Not cached the cache will crash if the default is a numpy array.
# The ppoperty calling this will need to cache itself and can safely do so since it has no parameters
def get_split_float_param(self, param, default=(0, 0)):
default = np.array(default) # type coersion in case the default is a tuple
raw = self.get_param(param, "")
parts = raw.split()
try:
if len(parts) == 0:
return default
elif len(parts) == 1:
a = float(parts[0])
return np.array([a, a])
else:
a = float(parts[0])
b = float(parts[1])
return np.array([a, b])
except (TypeError, ValueError):
return default
# not cached
def get_split_mm_param_as_px(self, param, default):
return self.get_split_float_param(param, default) * PIXELS_PER_MM
# returns an array of multiple space separated int values
@cache
def get_multiple_int_param(self, param, default="0"):
params = self.get_param(param, default).split(" ")
try:
params = [int(param) for param in params if param]
except (TypeError, ValueError):
return [default]
return params
def set_param(self, name, value):
# Sets a param on the node backing this element. Used by params dialog.
# After calling, this element is invalid due to caching and must be re-created to use the new value.
param = INKSTITCH_ATTRIBS[name]
self.node.set(param, str(value))
def remove_param(self, name):
param = INKSTITCH_ATTRIBS[name]
del self.node.attrib[param]
@cache
def _get_specified_style(self):
# We want to cache this, because it's quite expensive to generate.
return self.node.specified_style()
def get_style(self, style_name, default=None):
style = self._get_specified_style().get(style_name, default)
if style == 'none':
style = None
return style
@property
@cache
def stroke_scale(self):
# How wide is the stroke, after the transforms are applied?
#
# If the transform is just simple scaling that preserves the aspect ratio,
# then this is completely accurate. If there's uneven scaling or skewing,
# then the stroke is bent out of shape. We'll make an approximation based on
# the average scaling in the X and Y axes.
#
# Of course, transforms may also involve rotation, skewing, and translation.
# All except translation can affect how wide the stroke appears on the screen.
node_transform = inkex.transforms.Transform(get_node_transform(self.node))
# First, figure out the translation component of the transform. Using a zero
# vector completely cancels out the rotation, scale, and skew components.
zero = [0, 0]
zero = inkex.Transform.apply_to_point(node_transform, zero)
translate = Point(*zero)
# Next, see how the transform affects unit vectors in the X and Y axes. We
# need to subtract off the translation or it will affect the magnitude of
# the resulting vector, which we don't want.
unit_x = [1, 0]
unit_x = inkex.Transform.apply_to_point(node_transform, unit_x)
sx = (Point(*unit_x) - translate).length()
unit_y = [0, 1]
unit_y = inkex.Transform.apply_to_point(node_transform, unit_y)
sy = (Point(*unit_y) - translate).length()
# Take the average as a best guess.
node_scale = (sx + sy) / 2.0
return node_scale
@property
@cache
def stroke_width(self):
width = self.get_style("stroke-width", "1.0")
width = convert_length(width)
return width * self.stroke_scale
@property
@param('ties',
_('Allow lock stitches'),
tooltip=_('Tie thread at the beginning and/or end of this object. '
'Manual stitch will only add lock stitches if force lock stitched is checked.'),
type='dropdown',
# Ties: 0 = Both | 1 = Before | 2 = After | 3 = Neither
# L10N options to allow lock stitch before and after objects
options=[_("Both"), _("Before"), _("After"), _("Neither")],
default=0,
sort_index=50)
@cache
def ties(self):
return self.get_int_param("ties", 0)
@property
@param('force_lock_stitches',
_('Force lock stitches'),
tooltip=_('Sew lock stitches after sewing this element, '
'even if the distance to the next object is shorter than defined by the '
'minimum jump stitch length value in the Ink/Stitch preferences.'),
type='boolean',
default=False,
sort_index=51)
@cache
def force_lock_stitches(self):
return self.get_boolean_param('force_lock_stitches', False)
@property
@param('lock_start',
_('Tack stitch'),
tooltip=_('Tack down stitch type'),
type='combo',
default='half_stitch',
options=LOCK_DEFAULTS['start'],
sort_index=52)
def lock_start(self):
return self.get_param('lock_start', "half_stitch")
@property
@param('lock_custom_start',
_('Custom path'),
tooltip=_("Enter a custom path. For svg paths The last node will not be embroidered, but represents the first stitch of the element."),
type="string",
default="",
select_items=[('lock_start', 'custom')],
sort_index=53)
def lock_custom_start(self):
return self.get_param('lock_custom_start', '')
@property
@param('lock_start_scale_mm',
_('Scale tack stitch'),
tooltip=_('Set stitch length. A 1 in a custom path equals this values.'),
type='float',
unit="mm",
default=0.7,
select_items=[('lock_start', lock.id) for lock in LOCK_DEFAULTS['start'] if isinstance(lock, (AbsoluteLock, CustomLock))],
sort_index=54)
def lock_start_scale_mm(self):
return self.get_float_param('lock_start_scale_mm', 0.7)
@property
@param('lock_start_scale_percent',
_('Scale tack stitch'),
tooltip=_('Scale tack stitch by this percentage.'),
type='float',
unit="%",
default=100,
select_items=[('lock_start', lock.id) for lock in LOCK_DEFAULTS['start'] if isinstance(lock, (SVGLock, CustomLock))],
sort_index=54)
def lock_start_scale_percent(self):
return self.get_float_param('lock_start_scale_percent', 100)
@property
@param('lock_end',
_('Lock stitch'),
tooltip=_('Lock stitch type'),
type='combo',
default='half_stitch',
options=LOCK_DEFAULTS['end'],
sort_index=55)
def lock_end(self):
return self.get_param('lock_end', "half_stitch")
@property
@param('lock_custom_end',
_('Custom path'),
tooltip=_("Enter a custom path. For svg paths the first node will not be embroidered, but represents the last stitch of the element."),
type="string",
default="",
select_items=[('lock_end', 'custom')],
sort_index=56)
def lock_custom_end(self):
return self.get_param('lock_custom_end', '')
@property
@param('lock_end_scale_mm',
_('Scale lock stitch'),
tooltip=_('Set length of lock stitches (mm).'),
type='float',
unit="mm",
default=0.7,
select_items=[('lock_end', lock.id) for lock in LOCK_DEFAULTS['end'] if isinstance(lock, (AbsoluteLock, CustomLock))],
sort_index=57)
def lock_end_scale_mm(self):
return self.get_float_param('lock_end_scale_mm', 0.7)
@property
@param('lock_end_scale_percent',
_('Scale lock stitch'),
tooltip=_('Scale lock stitch by this percentage.'),
type='float',
unit="%",
default=100,
select_items=[('lock_end', lock.id) for lock in LOCK_DEFAULTS['end'] if isinstance(lock, (SVGLock, CustomLock))],
sort_index=57)
@cache
def lock_end_scale_percent(self):
return self.get_float_param('lock_end_scale_percent', 100)
@property
@param('trim_after',
_('Trim After'),
tooltip=_('Add a TRIM command after stitching this object.'),
type='boolean',
default=False,
sort_index=60)
def trim_after(self):
return self.get_boolean_param('trim_after', False)
@property
@param('stop_after',
_('Stop After'),
tooltip=_('Add a STOP command after stitching this object.'),
type='boolean',
default=False,
sort_index=60)
def stop_after(self):
return self.get_boolean_param('stop_after', False)
@property
def path(self):
# A CSP is a "cubic superpath".
#
# A "path" is a sequence of strung-together bezier curves.
#
# A "superpath" is a collection of paths that are all in one object.
#
# The "cubic" bit in "cubic superpath" is because the bezier curves
# inkscape uses involve cubic polynomials.
#
# Each path is a collection of tuples, each of the form:
#
# (control_before, point, control_after)
#
# A bezier curve segment is defined by an endpoint, a control point,
# a second control point, and a final endpoint. A path is a bunch of
# bezier curves strung together. One could represent a path as a set
# of four-tuples, but there would be redundancy because the ending
# point of one bezier is the starting point of the next. Instead, a
# path is a set of 3-tuples as shown above, and one must construct
# each bezier curve by taking the appropriate endpoints and control
# points. Bleh. It should be noted that a straight segment is
# represented by having the control point on each end equal to that
# end's point.
#
# In a path, each element in the 3-tuple is itself a tuple of (x, y).
# Tuples all the way down. Hasn't anyone heard of using classes?
if getattr(self.node, "get_path", None):
d = self.node.get_path()
else:
d = self.node.get("d", "")
if not d:
self.fatal(_("Object %(id)s has an empty 'd' attribute. Please delete this object from your document.") % dict(id=self.node.get("id")))
return inkex.Path(d).to_superpath()
@cache
def parse_path(self):
return apply_transforms(self.path, self.node)
@property
@cache
def paths(self):
return self.flatten(self.parse_path())
@property
def shape(self):
raise NotImplementedError("INTERNAL ERROR: %s must implement shape()", self.__class__)
@property
@cache
def commands(self):
return find_commands(self.node)
@cache
def get_commands(self, command):
return [c for c in self.commands if c.command == command]
@cache
def has_command(self, command):
return len(self.get_commands(command)) > 0
@cache
def get_command(self, command):
commands = self.get_commands(command)
if commands:
return commands[0]
else:
return None
def strip_control_points(self, subpath):
return [point for control_before, point, control_after in subpath]
def flatten(self, path):
"""approximate a path containing beziers with a series of points"""
path = deepcopy(path)
bezier.cspsubdiv(path, 0.1)
return [self.strip_control_points(subpath) for subpath in path]
def flatten_subpath(self, subpath):
path = [deepcopy(subpath)]
bezier.cspsubdiv(path, 0.1)
return self.strip_control_points(path[0])
@property
@cache
def lock_stitches(self):
lock_start = None
lock_end = None
# Ties: 0 = Both | 1 = Before | 2 = After | 3 = Neither
tie_modus = self.ties
force = self.force_lock_stitches
if tie_modus in [0, 1]:
lock_start = LockStitch('start', self.lock_start, scale_percent=self.lock_start_scale_percent, scale_absolute=self.lock_start_scale_mm)
if self.lock_start == "custom":
lock_start.set_path(self.lock_custom_start)
if tie_modus in [0, 2] or force:
lock_end = LockStitch('end', self.lock_end, scale_percent=self.lock_end_scale_percent, scale_absolute=self.lock_end_scale_mm)
if self.lock_end == "custom":
lock_end.set_path(self.lock_custom_end)
return lock_start, lock_end
def to_stitch_groups(self, last_patch):
raise NotImplementedError("%s must implement to_stitch_groups()" % self.__class__.__name__)
@debug.time
def _load_cached_stitch_groups(self, previous_stitch):
if not self.uses_previous_stitch():
# we don't care about the previous stitch
previous_stitch = None
cache_key = self.get_cache_key(previous_stitch)
stitch_groups = get_stitch_plan_cache().get(cache_key)
if stitch_groups:
debug.log(f"used cache for {self.node.get('id')} {self.node.get(INKSCAPE_LABEL)}")
else:
debug.log(f"did not use cache for {self.node.get('id')} {self.node.get(INKSCAPE_LABEL)}, key={cache_key}")
return stitch_groups
def uses_previous_stitch(self):
"""Returns True if the previous stitch can affect this Element's stitches.
This function may be overridden in a subclass.
"""
return False
@debug.time
def _save_cached_stitch_groups(self, stitch_groups, previous_stitch):
stitch_plan_cache = get_stitch_plan_cache()
cache_key = self.get_cache_key(previous_stitch)
if cache_key not in stitch_plan_cache:
stitch_plan_cache[cache_key] = stitch_groups
if previous_stitch is not None:
# Also store it with None as the previous stitch, so that it can be used next time
# if we don't care about the previous stitch
cache_key = self.get_cache_key(None)
if cache_key not in stitch_plan_cache:
stitch_plan_cache[cache_key] = stitch_groups
def get_params_and_values(self):
params = {}
for param in self.get_params():
params[param.name] = self.get_param(param.name, param.default)
return params
@cache
def _get_patterns_cache_key_data(self):
return get_patterns_cache_key_data(self.node)
@cache
def _get_guides_cache_key_data(self):
return get_marker_elements_cache_key_data(self.node, "guide-line")
def _get_gradient_cache_key_data(self):
gradient = {}
if hasattr(self, 'gradient') and self.gradient is not None:
gradient['stops'] = self.gradient.stop_offsets
gradient['orientation'] = [self.gradient.x1(), self.gradient.x2(), self.gradient.y1(), self.gradient.y2()]
gradient['styles'] = [(style['stop-color'], style['stop-opacity']) for style in self.gradient.stop_styles]
return gradient
def get_cache_key_data(self, previous_stitch):
return []
def get_cache_key(self, previous_stitch):
cache_key_generator = CacheKeyGenerator()
cache_key_generator.update(self.__class__.__name__)
cache_key_generator.update(self.get_params_and_values())
cache_key_generator.update(self.parse_path())
cache_key_generator.update(list(self._get_specified_style().items()))
cache_key_generator.update(self._get_gradient_cache_key_data())
cache_key_generator.update(previous_stitch)
cache_key_generator.update([(c.command, c.target_point) for c in self.commands])
cache_key_generator.update(self._get_patterns_cache_key_data())
cache_key_generator.update(self._get_guides_cache_key_data())
cache_key_generator.update(self.get_cache_key_data(previous_stitch))
cache_key = cache_key_generator.get_cache_key()
debug.log(f"cache key for {self.node.get('id')} {self.node.get(INKSCAPE_LABEL)} {previous_stitch}: {cache_key}")
return cache_key
def embroider(self, last_stitch_group):
debug.log(f"starting {self.node.get('id')} {self.node.get(INKSCAPE_LABEL)}")
with self.handle_unexpected_exceptions():
if last_stitch_group:
previous_stitch = last_stitch_group.stitches[-1]
else:
previous_stitch = None
stitch_groups = self._load_cached_stitch_groups(previous_stitch)
if not stitch_groups:
self.validate()
stitch_groups = self.to_stitch_groups(last_stitch_group)
apply_patterns(stitch_groups, self.node)
if stitch_groups:
stitch_groups[-1].trim_after = self.has_command("trim") or self.trim_after
stitch_groups[-1].stop_after = self.has_command("stop") or self.stop_after
self._save_cached_stitch_groups(stitch_groups, previous_stitch)
debug.log(f"ending {self.node.get('id')} {self.node.get(INKSCAPE_LABEL)}")
return stitch_groups
def fatal(self, message, point_to_troubleshoot=False):
label = self.node.get(INKSCAPE_LABEL)
id = self.node.get("id")
if label:
name = "%s (%s)" % (label, id)
else:
name = id
error_msg = f"{name}: {message}"
if point_to_troubleshoot:
error_msg += "\n\n%s" % _("Please run Extensions > Ink/Stitch > Troubleshoot > Troubleshoot objects. "
"This will show you the exact location of the problem.")
raise InkstitchException(error_msg)
@contextmanager
def handle_unexpected_exceptions(self):
try:
# This runs the code in the `with` body so that we can catch
# exceptions.
yield
except (InkstitchException, SystemExit, KeyboardInterrupt):
raise
except Exception:
if hasattr(sys, 'gettrace') and sys.gettrace():
# if we're debugging, let the exception bubble up
raise
raise InkstitchException(format_uncaught_exception())
def validation_errors(self):
"""Return a list of errors with this Element.
Validation errors will prevent the Element from being stitched.
Return value: an iterable or generator of instances of subclasses of ValidationError
"""
return []
def validation_warnings(self):
"""Return a list of warnings about this Element.
Validation warnings don't prevent the Element from being stitched but
the user should probably fix them anyway.
Return value: an iterable or generator of instances of subclasses of ValidationWarning
"""
return []
def is_valid(self):
# We have to iterate since it could be a generator.
for error in self.validation_errors():
return False
return True
def validate(self):
"""Print an error message and exit if this Element is invalid."""
for error in self.validation_errors():
# note that self.fatal() exits, so this only shows the first error
self.fatal(error.description, True)