Skip to content
Cannot retrieve contributors at this time
# -*- coding: utf-8 -*-
# Copyright (c) Vispy Development Team. All Rights Reserved.
# Distributed under the (new) BSD License. See LICENSE.txt for more info.
from __future__ import division
from .linear import STTransform, NullTransform
from .chain import ChainTransform
from ._util import TransformCache
from ...util.event import EventEmitter
import numpy as np
class TransformSystem(object):
"""TransformSystem encapsulates information about the coordinate
systems needed to draw a Visual.
Visual rendering operates in six coordinate systems:
* **Visual** - arbitrary local coordinate frame of the visual. Vertex
buffers used by the visual are usually specified in this coordinate
* **Scene** - This is an isometric coordinate system used mainly for
lighting calculations.
* **Document** - This coordinate system has units of _logical_ pixels, and
should usually represent the pixel coordinates of the canvas being drawn
to. Visuals use this coordinate system to make measurements for font
size, line width, and in general anything that is specified in physical
units (px, pt, mm, in, etc.). In most circumstances, this is exactly the
same as the canvas coordinate system.
* **Canvas** - This coordinate system represents the logical pixel
coordinates of the canvas. It has its origin in the top-left corner of
the canvas, and is typically the coordinate system that mouse and touch
events are reported in. Note that, by convention, _logical_ pixels
are not necessarily the same size as the _physical_ pixels in the
framebuffer that is being rendered to.
* **Framebuffer** - The buffer coordinate system has units of _physical_
pixels, and should usually represent the coordinates of the current
framebuffer (on the canvas or an FBO) being rendered to. Visuals use this
coordinate system primarily for antialiasing calculations. It is also the
coorinate system used by glFragCoord. In most cases,
this will have the same scale as the document and canvas coordinate
systems because the active framebuffer is the
back buffer of the canvas, and the canvas will have _logical_ and
_physical_ pixels of the same size. However, the scale may be different
in the case of high-resolution displays, or when rendering to an
off-screen framebuffer with different scaling or boundaries than the
* **Render** - This coordinate system is the obligatory system for
vertices returned by a vertex shader. It has coordinates (-1, -1) to
(1, 1) across the current glViewport. In OpenGL terminology, this is
called clip coordinates.
canvas : Canvas
The canvas being drawn to.
dpi : float
The dot-per-inch resolution of the document coordinate system. By
default this is set to the resolution of the canvas.
By default, TransformSystems are configured such that the document
coordinate system matches the logical pixels of the canvas,
1. To convert local vertex coordinates to normalized device coordinates in
the vertex shader, we first need a vertex shader that supports configurable
vec4 a_position;
void main() {
gl_Position = $transform(a_position);
Next, we supply the complete chain of transforms when drawing the visual:
def draw(tr_sys):
tr = tr_sys.get_full_transform()
self.program['transform'] = tr.shader_map()
self.program['a_position'] = self.vertex_buffer
2. Draw a line whose width is given in mm. To start, we need normal vectors
for each vertex, which tell us the direction the vertex should move in
order to set the line width::
vec4 a_position;
vec4 a_normal;
float u_line_width;
float u_dpi;
void main() {
// map vertex position and normal vector to the document cs
vec4 doc_pos = $visual_to_doc(a_position);
vec4 doc_normal = $visual_to_doc(a_position + a_normal) - doc_pos;
// Use DPI to convert mm line width to logical pixels
float px_width = (u_line_width / 25.4) * dpi;
// expand by line width
doc_pos += normalize(doc_normal) * px_width;
// finally, map the remainder of the way to normalized device
// coordinates.
gl_Position = $doc_to_render(a_position);
In this case, we need to access
the transforms independently, so ``get_full_transform()`` is not useful
def draw(tr_sys):
# Send two parts of the full transform separately
self.program['visual_to_doc'] = tr_sys.visual_to_doc.shader_map()
doc_to_render = (tr_sys.framebuffer_transform *
self.program['visual_to_doc'] = doc_to_render.shader_map()
self.program['u_line_width'] = self.line_width
self.program['u_dpi'] = tr_sys.dpi
self.program['a_position'] = self.vertex_buffer
self.program['a_normal'] = self.normal_buffer
3. Draw a triangle with antialiasing at the edge.
4. Using inverse transforms in the fragment shader
def __init__(self, canvas=None, dpi=None):
self.changed = EventEmitter(source=self, type='transform_changed')
self._canvas = None
self._fbo_bounds = None
self.canvas = canvas
self._cache = TransformCache()
self._dpi = dpi
self._mappings = {'ct0': None, 'ct1': None, 'ft0': None}
# Assign a ChainTransform for each step. This allows us to always
# return the same transform objects regardless of how the user
# configures the system.
self._visual_transform = ChainTransform([NullTransform()])
self._scene_transform = ChainTransform([NullTransform()])
self._document_transform = ChainTransform([NullTransform()])
self._canvas_transform = ChainTransform([STTransform(),
self._framebuffer_transform = ChainTransform([STTransform()])
for tr in (self._visual_transform, self._scene_transform,
self._document_transform, self._canvas_transform,
def _update_if_maps_changed(self, transform, map_key, new_maps):
"""Helper to store and check current (from, to) maps against new
ones being provided. The new mappings are only applied if a change
has occurred (and also stored in the current mappings).
if self._mappings[map_key] is None:
self._mappings[map_key] = new_maps
transform.set_mapping(new_maps[0], new_maps[1])
if np.any(self._mappings[map_key] != new_maps):
self._mappings[map_key] = new_maps
transform.set_mapping(new_maps[0], new_maps[1])
def configure(self, viewport=None, fbo_size=None, fbo_rect=None,
"""Automatically configure the TransformSystem:
* canvas_transform maps from the Canvas logical pixel
coordinate system to the framebuffer coordinate system, taking into
account the logical/physical pixel scale factor, current FBO
position, and y-axis inversion.
* framebuffer_transform maps from the current GL viewport on the
framebuffer coordinate system to clip coordinates (-1 to 1).
viewport : tuple or None
The GL viewport rectangle (x, y, w, h). If None, then it
is assumed to cover the entire canvas.
fbo_size : tuple or None
The size of the active FBO. If None, then it is assumed to have the
same size as the canvas's framebuffer.
fbo_rect : tuple or None
The position and size (x, y, w, h) of the FBO in the coordinate
system of the canvas's framebuffer. If None, then the bounds are
assumed to cover the entire active framebuffer.
canvas : Canvas instance
Optionally set the canvas for this TransformSystem. See the
`canvas` property.
# TODO: check that d2f and f2r transforms still contain a single
# STTransform (if the user has modified these, then auto-config should
# either fail or replace the transforms)
if canvas is not None:
self.canvas = canvas
canvas = self._canvas
if canvas is None:
raise RuntimeError("No canvas assigned to this TransformSystem.")
# By default, this should invert the y axis--canvas origin is in top
# left, whereas framebuffer origin is in bottom left.
map_from = [(0, 0), canvas.size]
map_to = [(0, canvas.physical_size[1]), (canvas.physical_size[0], 0)]
'ct1', np.array((map_from, map_to)))
if fbo_rect is None:
self._canvas_transform.transforms[0].scale = (1, 1, 1)
self._canvas_transform.transforms[0].translate = (0, 0, 0)
# Map into FBO coordinates
map_from = [(fbo_rect[0], fbo_rect[1]),
(fbo_rect[0] + fbo_rect[2], fbo_rect[1] + fbo_rect[3])]
map_to = [(0, 0), fbo_size]
'ct0', np.array((map_from, map_to)))
if viewport is None:
if fbo_size is None:
# viewport covers entire canvas
map_from = [(0, 0), canvas.physical_size]
# viewport covers entire FBO
map_from = [(0, 0), fbo_size]
map_from = [viewport[:2],
(viewport[0] + viewport[2], viewport[1] + viewport[3])]
map_to = [(-1, -1), (1, 1)]
'ft0', np.array((map_from, map_to)))
def canvas(self):
"""The Canvas being drawn to."""
return self._canvas
def canvas(self, canvas):
self._canvas = canvas
def dpi(self):
"""Physical resolution of the document coordinate system (dots per inch)."""
if self._dpi is None:
if self._canvas is None:
return None
return self.canvas.dpi
return self._dpi
def dpi(self, dpi):
assert dpi > 0
self._dpi = dpi
def visual_transform(self):
"""Transform mapping from visual local coordinate frame to scene coordinate frame."""
return self._visual_transform
def visual_transform(self, tr):
self._visual_transform.transforms = tr
def scene_transform(self):
"""Transform mapping from scene coordinate frame to document coordinate frame."""
return self._scene_transform
def scene_transform(self, tr):
self._scene_transform.transforms = tr
def document_transform(self):
"""Transform mapping from document coordinate frame to the framebuffer (physical pixel) coordinate frame."""
return self._document_transform
def document_transform(self, tr):
self._document_transform.transforms = tr
def canvas_transform(self):
"""Transform mapping from canvas coordinate frame to framebuffer coordinate frame."""
return self._canvas_transform
def canvas_transform(self, tr):
self._canvas_transform.transforms = tr
def framebuffer_transform(self):
"""Transform mapping from pixel coordinate frame to rendering coordinate frame."""
return self._framebuffer_transform
def framebuffer_transform(self, tr):
self._framebuffer_transform.transforms = tr
def get_transform(self, map_from='visual', map_to='render'):
"""Return a transform mapping between any two coordinate systems.
map_from : str
The starting coordinate system to map from. Must be one of: visual,
scene, document, canvas, framebuffer, or render.
map_to : str
The ending coordinate system to map to. Must be one of: visual,
scene, document, canvas, framebuffer, or render.
tr = ['visual', 'scene', 'document', 'canvas', 'framebuffer', 'render']
ifrom = tr.index(map_from)
ito = tr.index(map_to)
if ifrom < ito:
trs = [getattr(self, '_' + t + '_transform')
for t in tr[ifrom:ito]][::-1]
trs = [getattr(self, '_' + t + '_transform').inverse
for t in tr[ito:ifrom]]
return self._cache.get(trs)
def pixel_scale(self):
tr = self._canvas_transform
return (, 0)) -, 0)))[0]