Working with GXL and RSF

1
Working with GXL and RSF

• Jeff Michaud
• University of Victoria
• CASCON 2000

2
RSF and GXL The Formats

• Rigi Standard Format
• Nodes, arcs, and attributes
• Flat text, unordered, unorganized
• Meta model stored externally
• One large file

• Graph EXchange Language
• Nodes, edges, rels, and
• attributes
• XML (hierarchical structure)
• Graphs can contain graphs
• Meta model expressed within
  
• file
• Possibility to effectively split
• graph into multiple files
• Other flexible features

3
The Exchange Formats - RSF

• n!type 100!A Box
• n!colour 100!A Yellow
• n!type 200!B Rect
• n!colour 200!B Orange
• a!connect 100!300 200!B

Id 100 Name A Type Box Colour - Yellow
Id - 300
Id 200 Name B Type Rectangle Colour Red
4
The Exchange Formats - GXL

• ltgxlgt
• ltgraph id1gt
• ltnode id100 typeBoxgt
• ltattr namecolourgt
• ltstringgtYellowlt/stringgt
• lt/attrgt
• lt/nodegt
• ltnode id200 typeRectgt
• ltattr namecolourgt
• ltstringgtOrangelt/stringgt
• lt/attrgt
• lt/nodegt
• ltedge id300 typeconnect from100
  to200gt
• lt/graphgt
• lt/gxlgt

Id 100 Name A Type Box Colour - Yellow
Id - 300
Id 200 Name B Type Rectangle Colour Red
5
RSF to GXL

• implemented with Java and DOM
• successful
• difficulties
• first time programming with XML
• most libraries only supported reading XML at the
  time
• speed issues with large files
• attributes can appear anywhere in RSF

6
GXL to RSF

• first attempt XSLT
• FAILURE !!!
• unacceptable performance really slow with
  medium to large files
• required to search through GXL when composing
  arcs
• second attempt Java and DOM
• SUCCESS !!!
• memory problems with large files
• no GXL importing tools yet
• new DTDs every month
• more robust design for conversion process
• intermediate in memory model

7
Without a meta-model

• difficult to understand what the GXL from other
  sources would contain
• what each node and edge type will mean
• without the knowledge of what other meta-models
  would be, assumptions (incorrect ones) were made
  about how to convert the GXL
• Question How many meta models do we need?
• One for each conversion path?
• One for each domain (language, granularity)?

8
Conversion questions (and answers?)

• What should be done with rels?
• Interpret them as edges?
• What should be done with the Atomic data types?
• Currently are ignored on the way to RSF and types
  are determined on the way back to GXL.
• What should be done with the Composite data
  types?
• With a convention, composites are stored in RSF
  and can thus conversions back to GXL will retain
  composites. Nested composites not currently
  supported.
• Should 100 round trip be a requirement? Is it
  feasible?

9
Concluding thoughts on working with GXL

• really liked working with XML (and GXL)
• easy to read
• numerous tools (especially in Java)
• in memory model a good idea (perhaps more
  resistant to changes in DTD )
• blowup may not be an issue when converting from
  structured RSF