HN Explorer

Ask HN: What is your favorite application of Graphviz?

10 points by calcnerd256 ↗ HN
How do you use Graphviz? What have you written that uses it on the backend? What is your favorite tool that you use that uses it? What are some ideas you haven't gotten around to implementing that would use it?

8 comments

[ 2.7 ms ] story [ 28.0 ms ] thread
makecheck
I use it indirectly from Doxygen regularly. I've also had data in custom formats that is naturally graph-like, and it was trivial to produce visualizers by writing scripts to convert the custom format into "dot" syntax.

One idea I have is to use "dot" to give auto-layout capabilities to a GUI that has a simple canvas. Right now, any data set that wasn't created by a GUI user (e.g. automatically generated) looks ugly, because it has no useful layout information. Since "dot" knows how to do layout, it would seem possible to run it in the background, parse the results, and use the node positions to make the default canvas look much more intelligent.

thristian
My most useful GraphViz hack so far has been a script that reads the schema description out of a PostgreSQL database and turns it into a vaguely Access/SQL Server-like database diagram (a node for each table, arrows representing foreign-key relations).

Someday I'd like to set up a script that takes the query-plan returned by PostgreSQL's EXPLAIN statement and turns it into a diagram with a node for each operation and arrows for dependencies, where the arrow width is proportional to the number of rows involved (again, much like SQL Server's query plan results). This would make it easier to spot the hotspots in a complex plan.

My favourite GraphViz tool would probably be XDot:

    http://code.google.com/p/jrfonseca/wiki/XDot
Fast, interactive viewing of GraphViz source files without having to convert them to PNG every time, and can be embedded in your own Python/GTK+ apps.
hippee-lee
Would you mind sharing your Script for parsing the db schema?
thristian
I'd like to, but I wrote it on company time and hence don't actually have the right to distribute it.

I will say that the trickiest bits were extracting the exact information from the database, and figuring out the how to make arrows point to and from individual fields, instead of from the table in general.

Getting tables and fields from PostgreSQL:

    select 
        table_schema,
        table_schema || '.' || table_name as qualified_table, 
        column_name, 
        data_type 
    from information_schema.columns 
    order by table_schema, table_name, ordinal_position;
Getting foreign-key information from PostgreSQL:

    select
        kcu.table_schema,
        kcu.table_schema || '.' || kcu.table_name as qualified_table,
        kcu.column_name,
        ccu.table_schema as referenced_table_schema,
        ccu.table_schema || '.' || ccu.table_name 
            as referenced_qualified_table,
        ccu.column_name as referenced_column_name
    from information_schema.constraint_column_usage as ccu
        join information_schema.key_column_usage as kcu
        using (constraint_catalog, constraint_schema, constraint_name)
    where ccu.table_schema != kcu.table_schema 
        or ccu.table_name != kcu.table_name;
As for the DOT output, I wound up using their HTML table support, generating a unique name for each cell in the table (port="blah" on each td element) and defining the foreign key arrows as going from table1:field1:e to table2:field2:w when table1.field1 is a foreign key referencing table2.field2.
hippee-lee
I am using mysql. Maybe I can figure something out and post it back here ...

Thanks for the hints.

ananthrk
Did the same thing (plotting tables and relation betweeen them) for the C# DataSet object.
jbr
I just ran into http://diagrammr.com/ tonight. Pretty sweet little mini-app that looks like it's built on graphviz/dot.
hippee-lee
We had to write an awk script for a class a while ago that used the output from du for a given folder. The output from the script was a dot file with the directory structure and memory used for each node below the given directory argument. Mine was pretty simple but one student figured out how to make each node look like a folder and color coded the nodes based on the amout of space used.