MS RFC 104: Replace BITMAP labels with embedded Truetype font¶
1. The Current Situation¶
MapServer supports embedded BITMAP fonts for the simple rendering of text without need for external font files or fontsets. These are activated by using the TINY, SMALL, MEDIUM, LARGE and HUGE keywords in the LABEL block, but have a number of limitations:
- They are only supported by the AGG renderer. The Cairo and KML renderers will bailout if they are supplied with a bitmap font. GD was the other only renderer to support them, it has however been removed in the 7.0 release.
- They only support the ASCII character set. Accented letters or common symbols are either ignored or replaced with a stopgap character
- They do not support rotation. FOLLOW labels will result in an error, while explicitly rotated or AUTO labels will be incorrectly rendered horizontally.
- The quality of their rendering is not up to par with today’s cartographic rendering expectations.
They do however have the following advantages:
- They require no dependency on external font files, and therefore do not require creating or maintaining a fontset file. As such they are ideal for rapid mapfile prototyping.
- They do not go through freetype for rendering and thus are extremely fast to render (which seems only fair given their limitations)
1.1 Upcoming Evolution¶
With the arrival of MS RFC 98: Label/Text Rendering Overhaul and MS RFC 99: Remove support for GD renderer, maintaining the support for bitmap fonts requires quite a number of warts in the code. Given this and the other limitations enumerated beforehand, this RFC proposes the removal of the BITMAP font support throughout the MapServer codebase.
To maintain the ability to rapidly prototype mapfiles and/or to not have to rely on external truetype fonts and fontsets, they will however be replaced by an embedded truetype font.
2. Embedded Truetype Font¶
A Truetype font will be encoded and included as a binary blob inside the mapserver library. It will be used whenever no specific font has been supplied by the mapfile user, or as a backwards compatibility fallback when bitmap fonts are requested.
- Embedded truetype fonts will be automatically supported by all renderers
- Embedded truetype fonts will cover a much larger character set
- Embedded truetype fonts will support rotation and FOLLOW labels
- Code will be cleaned up and easier to maintain
- Rendering will be marginally slower than with bitmap fonts
- Due to the graphic quality of the rendered text, image sizes will be slightly larger than with bitmap fonts (due to the antialiasing).
2.2 Backwards Compatibility¶
Rendered images when using embedded fonts will differ compared to the previous renderings with bitmap fonts. Aside from this point, no backwards incompatibility is to be expected: The bitmap keyword and label type will be ignored, and the TINY, SMALL, MEDIUM, LARGE and HUGE keyboard will be mapped to integer font sizes of matching visual height.
2.3 Performance Implications¶
Might be noticeable:
- Slower rendering of bitmap glyphs
- Larger image sizes
2.4 Future Improvements¶
These are not planned immediately, but could be investigated in the future depending on interest or funding:
- Choosing at compile-time which truetype font should be embedded
- Embedding multiple fonts
- Allowing aliased glyph rendering from freetype to minimize image size
3. Implementation Details¶
3.1 Affected files¶
- mapfile.c/h: parser, backwards compatibility
- mapagg.cpp, maplabel.c, textlayout.c, others...: bitmap support removal
Bitmap related functions could be extended to provide a warning the feature has been removed. No other changes should be required due to the transparent backwards compatibility handling.
3.3 Tracking Issue¶
The truetype file we embed shall have a licence that allows such usage, and a glyph coverage supporting most used languages while not being too large to limit memory consumption. Unless decided otherwise during the comment period, the chosen font will be the DejaVuSansCondensed.ttf truetype font commonly supplied with Linux distros.
5. Voting History¶
+1 from SteveL, ThomasB, PericlesN, TamasS, TomK, StephanM, YewondwosssenA, MikeS, DanielM and StephenW