std::list can have a throwing move constructor.
std::vector of such lists makes copies when growing its storage array,
it doesn't move them.
render_thunk_list is noncopyable (because render_thunk is noncopyable),
and so can't be stored in std::vector in some STL implementations.
Wrapping render_thunk in std::unique_ptr is one extra allocation per
element, with no purpose. The somewhat costly xyz_render_thunk move
constructor is only called once upon insertion, regardless of whether
we're emplacing render_thunk or unique_ptr.
- refs #3327
Replace member variant of placement-type implementations with plain
union. The active implementation is chosen at construction time like
before.
Make placement-type implementation classes virtual to allow invoking
the active union member through a base class pointer.
- this reverts render_thunk_extractor visitation by text/shield
symbolizer back to how it works in master, only amended to avoid
passing temporaries to text_symbolizer_helper
- use abstract class render_thunk_list_dispatch for the callback
instead of template parameter render_thunks
- add class scoped_glyph_positions_offset for temporary shifting
of glyph positions
- remove render_offset_placements
This update is in Ref #2649. It provides an update to markers and their processing such that:
* All shared_ptrs are no longer required around markers
* Markers are now const
* Added new gray data types adding those to the variants and updating all the code necessary for them
* Added basic SSE to the image compare method, (only for RGBA) must be enabled with the -DSSE_MATH flag this is not yet put into the build process in any location.
* Fixed the resulting image for some TIFF visual tests, most likely they were incorrect due to fixes in TIFF reader
* Added some MAPNIK_DECL where necessary to grid rendering.
* Added support for more data types in GDAL plugin with grayscale images.
* Added views for all the new gray data types
* Updated python bindings for new gray data types.
Ref #2681
* Changes all agg_renderers to use a image_data_any variant (only image_data_rgba8 is implemented currently)
* Changes the marker and marker_cache to use image_data_any images
* Changes the symbolizers so that they must be aware of the source data type they are attempting to render and the render type that is expected to be rendered into.
* Moves many utilities into image_utils, that were previously in image_32.
The kicker is that this still isn't working perfectly yet, but I am commiting so I don't have tears in case everything is lost on my computer.
Ref #2633
image_data_32 -> image_data_rgba8
image_data_8 -> image_data_gray8
image_data_16 -> image_data_gray16
image_data_float32 -> image_data_gray32f
NOTE: currently image_data_gray16::pixel_type = std::int16_t (signed 16 bit int) to match GDAL
should we support unsigned types?
