MS RFC 77: Support for Multiple Label Objects Within a Class¶
Steve Lime, Thomas Bonfort
sdlime at comcast.net,
Adopted on 2011-12-15
This RFC proposes allowing for the definition of multiple label objects within a class. There are a number of use cases where this would be valuable.
Adding text below a label with different styling, for example an alternate name shown in italics.
Labeling roads with multiple shields and road numbers.
Supporting complex meteorlogical rendering where many values are placed around a point.
Presently only very limited support for these use cases is available. Basically you have to draw layers multiple times and place each piece of text separately. This is inefficient and does not allow for group label collision avoidance.
2. Proposed Technical Change¶
2.1 Core Object Changes¶
Class object would need to support multiple label objects in much the same way as the they support multiple styles. In fact, implementation would follow the styleObj pattern.
A key requirement is that the extra labels be self-contained, that is, they need to allow for the definition of label text and conditional display within the labelObj. To do this we propose adding the following attributes to labelObj's:
These attributes behave just like their classObj equivalents. In addition a labelObj will be extended with several attributes that will be used to hold per-feature values in the context of the label cache. They are:
labeltext (string to be drawn)
labelpoly (bounding box shapeObj)
A label's labeltext will be set according to the following order:
label text (new for this RFC)
A NULL labeltext value means that label will not be rendered.
The labelCacheMemberObj will also require change. First, it needs to support multiple labelObj's (see point and polygon discussion below) just like the styles attribute currently supported. A label counter will be added as well. Because the text to be drawn will be carried in a label the text attribute can be removed.
2.2 Label Rendering Changes¶
2.2.1 Point and Polygon Labels¶
The labeling of point and polygon features is done based on a single computed label point. Multiple labels within a classObj will be applied to a label point as a group, that is, all label text must be placed successfully or the entire group will not be drawn. In this case a new function, msAddLabelGroup() is available to add all labels from a class to the label cache as a single element. The bounding shapes for a group of labels will be considered as one for collision avoidance computations.
Positioning of the multiple labels will be done as always- relative to the label point using label->position and/or label->offset.
In the case of a label group label priority will be taken from the first label defined in a class. Scale thresholds will be taken into account when a labels labeltext value is assigned (outside of the cache population code). A label that is out of scale will set labeltext=NULL and will not be drawn although others in the group could be.
Label cache collision functions will have to be updated to look a multiple labels, especially the code that looks for duplicate text (label->mindistance).
2.2.2 Line Labels¶
Line labeling is another story. Lines often are used to derive curved text, repeating labels and sometimes both. In this case a single feature might contribute dozens of labels to the cache. It is impractical to handle these as we would a point or polygon with a single label point. A for more common use will be to label a line with multiple names or a combination of a name and a marker/text combination. So, for line labels the approach will be to simply use the existing cache faciliities but with an added loop to process multiple labels if available.
It might be possible in the future to use the label grouping described above in certain instances but that is outside the scope of this work.
Class objects will need support for labels similar to what is in place for styles. Presently a classObj supports the following methods: getStyle, insertStyle, removeStyle, moveStyleUp and moveStyleDown. We propose adding label equivalents for the first three (label order is not nearly as import as style order).
It is not quite clear how classes label is defined/manipulated from within MapScript. In SWIG/MapScript labels are marked as immutable. Further work needs to be done to understand if this represents a regression.
3. Implementation Details¶
3.1. Files Affected¶
Change lis limited primarily to:
mapserver.h: core structure changes
mapfile.c: read/write/initialize changes
mapdraw.c: significant changes to msDrawShape() and msDrawLabelCache()
maplabel.c: changes to msAddLabel() and addition of msAddLabelGroup()
maplayer.c: changes to msLayerWhichItems() to account for label text and expressions
maputil.c: changes to msBindLayerToShape() and msShapeGetAnnotation()
Various other files see small changes with the change in classObj from one label to many.
3.2 Documentation Changes¶
Documentation for classObj (labels 0 .. n) and labelObj (text and expression) will need update. MapScript documentation will need changes to document new methods and the change in object hierarchy between classes and labels. Mapfile XML file must be changed as well. A number of test cases will added to the MapServer regression test suite- all existing labeling tests should continue to be useful.
3.3 Bug ID¶
#4127. Work is underway in a sandbox (http://svn.osgeo.org/mapserver/sandbox/sdlime/rfc-77/).
4 Configuration Examples¶
Adding optional text below other text:
LABELITEM 'item1' CLASS ... LABEL # always displayed using LAYER:LABELITEM as a source for text FONT 'arial' ... END LABEL # conditionally displayed using LABEL:TEXT as a source for text EXPRESSION ('[item2]' ~ '.') # item2 has something in it FONT 'arial-italic' TEXT '[item2]' OFFSET 0 15 ... END END
5. Backwards Compatibility¶
At the mapfile level there are no issues, existing mapfiles should run without any change. It is possible that this could introduce regressions within MapScript although the scope is not fully known.
5.1 Relationship to Other Outstanding Label Bugs¶
#1355 - None, test is valid w/grouped labels too.
#2866 - None.
#2981 - None.
#3009 - None.
#3044 - This is already supported (ticket can be closed).
#3335 - Styles are supported for each additional label.
#3675 - This will be addressed by this RFC. Label content will be defined by the new TEXT attribute.
Although beyond the scope of this work it seems that some interesting work related to the relative placement of grouped labels might be possible. For example, it could be useful to define positions relative to the previous label instead of relative to a label point.
7. Voting history¶
Adopted on 12-15-2011 with +1 from SteveL, DanM, JeffM, ThomasB, SteveW, HowardW.