Re: d3js chord diagrams

[Ricardo Wurmus]

Ludovic Courtès <address@hidden> writes:

> Ricardo Wurmus <address@hidden> skribis:
>
>> I’ve built something:
>>
>>     http://elephly.net/graph.html
>
> Awesome!  Much better than staring at a static graph in Evince.
>
>> I’ve tried earlier to build a force-directed graph to visualise the
>> package dependency graph, but I had to realise that a force-directed
>> graph with the number of links and nodes that is common in software
>> dependencies is visually indistinguishable from a cat’s hair ball.  In
>> contrast I find the chord diagram to be much clearer.
>
> Are there other types of visualizations supported by d3 that would help?

Yes.  A force-directed graph could be made clearer by applying force
bundling as is done here: <https://github.com/upphiminn/d3.ForceBundle>.

We could map the graph to a collapsible tree such as this
<http://codepen.io/mikefab/full/IDdts/>.  One disadvantage of using a
tree is that we can no longer visually unify shared inputs.

We could also provide a variant of the chord diagram by bundling edges
as is done here:
<http://mbostock.github.io/d3/talk/20111116/bundle.html>.  We’d lose the
relative “importance” of a package, which in a chord diagram can be seen
in the radial size of a segment, but dependencies might be clearer.

(Graphviz-style representations are more difficult, because the location
of the nodes needs to be laid out first, whereas in d3 no
pre-computation takes place.)

So, lots of things that a dedicated hacker could implement :)

>> I turned this into a backend for “guix graph”, so that you can generate
>> There are a couple of implementation issues that I’d like to get input
>> on before submitting a proper patch.  This depends on “d3.v3.js”, which
>> I’ve downloaded from http://d3js.org/d3.v3.js.  (This is not minified
>> and thus rather large.)  In my current implementation the contents of
>> this file are embedded in the report, because we don’t serve this file
>> at a well-known location.  Is this okay?  We probably could use file://
>> links, but that requires knowledge about where this file is located.
>
> What about having a fixed-output derivation (aka. ‘origin’) to download
> that file from a stable, versioned URL?
>
>> If it is okay to bundle d3js with Guix, where should it be installed?
>> How can graph.scm know about the location of this file after
>> installation?  Should I add the path to d3.v3.js to
>> “guix/config.scm.in”?
>
> An option is to download it to the store on demand.  The downside, of
> course, is that it would require network access.
>
> If the .js files were to be installed, I agree with using PREFIX/lib/js
> as Pjotr suggests.

Does this mean that the generated HTML document would have to be a gexp
referencing the file in the store?  (I’m not sure how to implement
this, but with a pointer or two I could give it a try anyway.)

>> Should this go to “$out/share/guix/”?  How would these files be found
>> when using “./pre-inst-env”?
>
> Well, if it’s just the small graph.js, you could store it in moduledir,
> like we do for patches.  This is a hack, but the advantage is that we
> can just use (search-path %load-path …) to find them, regardless of
> whether we’re using ./pre-inst-env or not.

Okay.

> Otherwise we could add a new GUIX_JAVASCRIPT_PATH env. var. and
> corresponding ‘search-javascript-file’ procedure.

I like this a little less than the “search-path” option.

~~ Ricardo