3D Representations for Software Visualization

Andrian Marcus, Louis Feng, Jonathan I. MaleticSoftvis 2003 · DOI

TL;DR

This paper introduces sv3D, a novel 3D framework for visualizing large software systems that extends the SeeSoft metaphor. By leveraging a third dimension and enhanced manipulation techniques, sv3D aims to represent higher-level decisional data more effectively than traditional 2D methods, aiding in software comprehension and reengineering tasks.

What Problem Does This Paper Address?

Visualizing large-scale software systems presents a significant challenge for developers, especially during maintenance and reengineering. Existing 2D visualization techniques, while useful, often struggle with scalability, representing complex relationships, and handling multiple levels of abstraction. The paper highlights the limitations of popular methods like SeeSoft, which, despite its strengths in direct mapping to source code, faces issues with representing hierarchical data and information overload due to the constraints of 2D space. The authors aim to overcome these limitations by exploring the potential of 3D representations.

Methods

The core of this work is the sv3D framework, which builds upon the SeeSoft metaphor but introduces several key enhancements:

  • 3D Renderings: sv3D creates three-dimensional visualizations of the raw software data, moving beyond the 2D limitations of SeeSoft.
  • Enhanced Mapping: It allows various software system artifacts and their attributes to be mapped to 3D metaphors at different abstraction levels. This includes mapping attributes like nesting level to cylinder height and control structures to cylinder color.
  • Improved User Interaction: The framework incorporates advanced manipulation techniques, such as zooming, panning, rotation, scaling, and transparency control, to help users navigate and understand complex 3D visualizations.
  • Extensible Framework: sv3D is designed as an extensible framework, allowing for easy integration with different analysis tools and customization of visual elements and mappings. It uses an XML input format for data.
  • Leveraging Existing Toolkits: The implementation utilizes Qt for the GUI and Open Inventor for rendering, ensuring portability and efficiency.

Datasets

The paper does not explicitly mention specific large-scale software datasets used for training or evaluation. Instead, it uses a small mailing system example consisting of 30 C++ source code files (approximately 4000 LOC) to illustrate the visualization capabilities of sv3D. This example is used to demonstrate how different attributes (nesting level, control structures) can be mapped to visual elements like color and height.

Key Insights and Ideas

  • 3D for Higher-Level Data: The paper argues that 3D representations are better suited for visualizing higher-dimensional and more complex decisional data in software systems, which is often difficult to convey effectively in 2D.
  • Balancing Information and Overload: A key challenge in 3D visualization is presenting a lot of information without overwhelming the user. sv3D attempts to address this through careful design of visual metaphors and interactive manipulation techniques like transparency and elevation.
  • Importance of User Interaction: The authors emphasize that the effectiveness of a 3D visualization heavily relies on intuitive and powerful user interaction mechanisms. sv3D's focus on manipulation tools is crucial for making the 3D space navigable.
  • Flexibility in Mapping: The framework's ability to define custom mappings between data attributes and visual elements is a significant advantage, allowing users to tailor visualizations to specific tasks and data types.
  • Complementary Views: The paper suggests that different views (e.g., one mapping color to nesting level and another to control structures) can be used in conjunction to provide a more comprehensive understanding of the software.

Key Results

The paper presents qualitative results and demonstrations of the sv3D framework. It shows how sv3D can:

  • Visualize a system with multiple attributes (nesting level, control structures) simultaneously using color and height.
  • Use transparency and elevation to mitigate occlusion issues in 3D visualizations.
  • Support user tasks like overview, zoom, filter, details-on-demand, and relating items through its interactive features.
  • The authors claim that sv3D performs "exceptionally well in representing up to 40-50 KLOC," with ongoing work to improve performance for larger systems.

Limitations and Future Work

The authors acknowledge several limitations and outline future directions:

  • Scalability: While sv3D aims to address scalability, performance can degrade for systems larger than 40-50 KLOC. Improving rendering efficiency for larger systems is a key future goal.
  • Abstraction Levels: The current version primarily visualizes source code files. Future work includes supporting higher abstraction levels, such as functions, classes, or packages.
  • Mapping Flexibility: The paper plans to support variable numbers of edges for poly cylinders and different textures. The mapping of cylinder position within a container is also intended to be more flexible.
  • Hierarchical Data: Representing hierarchical data and diagrammatic visualizations (like UML class diagrams) is an area for future development, potentially through connecting containers with edges.
  • Stereoscopic Visualization: A stereoscopic version (sv3Ds) is being implemented to enhance depth perception.
  • User Studies: The authors recognize the need for controlled user studies to rigorously assess the advantages and disadvantages of sv3D and its efficiency compared to other methods.
  • Extraction and Querying: These features are planned for future implementation.

Summary updated

BibTeX
@inproceedings{marcus2003d,
  author = {Andrian Marcus and Louis Feng and Jonathan I. Maletic},
  title = {{3D Representations for Software Visualization}},
  booktitle = {Softvis},
  year = {2003},
  doi = {10.1145/774833.774837},
}
Figure