By default, alpha premultiplying should be done by the initial rendering
(eg quads) and not the final post processing pass.
The default post processing pass expects premultiplied color values and
will not do this operation itself.
This also makes it possible to call `solveConstraints_` with a duration of 0,
which means the target values are modified straight away without triggering
an animation or leaving the current flow.
Previously the formula for the resolution constraint allowed going way past
the minimum zoom.
Also adjusted the center constraint to avoid a zigzag effect when going out
of resolution bounds.
The DragPan, KeyboardPan and DragZoom interactions now make sure to
animate to a valid center/resolution target to avoid
a chained "resolve" animation which looks weird.
The `View.fit` method was also fixed to use this.
API changes:
* (breaking) the `rotate` method is gone
* the `adjustRotation`, `adjustResolution` and `adjustZoom` methods are now
available and allow using an anchor.
This means interactions do not have to do the anchor computation themselves
and this also fix anchor computation when constraints must be applied.
This introduces a breaking change.
This options replaces the various `constrainResolution` options on interactions
and the `fit` method.
Since constraints are now the responsibility of the View, the fact that intermediary
zoom levels are allowed or not is now set in the view options.
By default, the view resolution is unconstrained.
This will help making sure that the view will come back to a "rested" state
once the interactions are over.
Interactions no longer need to handle the animation back to a rested state,
they simply call `endInteraction` with the desired duration and direction.
This commit aims to simplify the computation of zoom and resolution in the
View class.
Previously zoom levels and resolution computations were mixed in different places,
ie resolution constraints, initial values, etc.
Now the View class only has the `getZoomForResolution` and `getResolutionForZoom`
methods to convert from one system to another.
Other than that, most computations use the resolution system internally.
The `constrainResolution` method also does not exist anymore, and is replaced
by `getValidResolution` and `getValidZoomLevel` public methods.
This introduces a breaking change:
The `extent` view option now constrains the whole viewport and not just the
view center.
The option `constrainOnlyCenter` was added to keep the previous behaviour.
Constraining the whole viewport and not only the view center means
that the center and resolution constraints must be applied with a knowledge of
the viewport size.
The `getValidZoomLevel` apply the current resolution constraint to return
a value that is guaranteed valid.
This is used for interactions & controls which need a target value to work:
the +/- buttons, the zoom clider, the dragbox zoom and the mouse wheel zoom.
The view now has targetCenter, targetRotation and targetResolution members.
These hold the new values given by set* methods. The actual view parameters are then changed by
calling `applyParameters_`.
Things done:
- stay closer to the original rendering in canvas
- get rid of the intermediary blur step
- use the new callbacks in the webgl points renderer
- premultiply alpha in the last postprocessing step
Now the shader and program caches are simply arrays of native WebGL created objects.
The WebGLHelper simply takes the sources of the frag and vert shader and produces a program.
This removes 2 classes & reduces the general verbosity of the API.
Also a `getShaderCompilationErrors` was added on `WebGLHelper` to help debug GLSL errors.
Olivier Guyot