In OpenLayers 2.12, the backbuffer for each layer was placed right below
the layer itself. With animated zooming, that was changed so backbuffers
are all placed below the base layer. As it turned out, this is only good
for certain kinds of layers (e.g. street labels), so the default now is
to place the backbuffer right below its layer again. But the
transitionEffect option can now have a new value, 'map-resize', which will
cause the backbuffer to be placed below the base layer.
Since we don't rely on the DOM order for layers, we also cannot rely on it
for backbuffers. This change simply adds a sensible z-index to the
backbuffer container.
When an application zooms the map programmatically right after
initialization, there will be an empty backbuffer. This should be avoided,
because it adds an unnecessary DOM element.
The tile.onLoadEnd handler registers a transitionend listener on the image
of the last loaded tile. But if loading was aborted, there will be no
image. The new logic registers the transitionend event on the last child
of the layer div, which is not necessarily the tile we're handling loadend
for. When the backbuffer is empty (i.e. no tile was loaded at the time the
backbuffer was created), it will be removed immediately.
There are still tests failing, but this change addresses some of the new failures after 4b163e0482 (which is an awesome improvement despite the test failures :).
This change introduces a new 'replace' mode for tile transitions: when the
resolution does not change, which happens when mergeNewParams is called,
the tile will be marked with the .olTileReplace class. If this class sets
the tile's imgDiv display to 'none', the backbuffer for the tile will
immediately be removed when the tile is loaded.
We now reuse tile images by maintaining a cache of image elements with a
simplified LRU expiry policy (by order, not by timestamp). The tile queue
is bypassed for images that are available in the cache, so they can be
rendered immediately. And the tile queue itself loads more than just one
image at a time now (2 per layer url).
This avoids issues with Elements renderers, where the nodeFactory method
cannot find nodes while the map viewport is transitioning from the mapDiv
to the googleControl.
Simple and effective: As soon as a map has a Google layer, the whole map viewport is added as control to the GMap. As soon as no Google layer is visible on the map any more, the map viewport is appended to the map container again. With this change, OpenLayers strictly limits its GMaps integration to the GMaps API.
Also note that there are no css overrides for the attribution any more. Instead, controls can now be conditionally positioned differently for Google layer by using the .olForeignContainer selector.
Using the Timeline tab of the Chrome Developer Tools, no significant
difference of Paint events can be observed when requestAnimationFrame is
used. So I agree with @elemoine that there is no need to introduce
asynchronous behavior here.
This also fixes an issue that has gone unnoticed for a while: the grid did
not cover the bottom of the map viewport, but instead covered an invisible
area above the top of the map viewport.
Now we also do not use deltas for shiftRow and shiftColumn. Some
refactoring was done so we do not need different calculateGridLayout
methods for layers with top-left and bottom-left tile origin.
TODO: With this commit, ArcGisCache and KaMap layers are broken.
This is because the threshold used for deciding when a column or row is shifted is too far to the bottom right. A tiny fix, but effective. A new test makes sure that we don't shift columns more than necessary when the layer is dragged.
