View Dependencies as a Tree

[schettino72]

This issue was first raised in this discussion

[ankostis]

Did you have something like this in mind?

I.e., try with project's dodo.py the following:

  doit list --graph
  doit list --graph epydoc package

[schettino72]

Cool. I am glad to see some work being done on this :)

• I think better leave the list command as it is now. And put graph as a separate command, it will probably grow to require its own parameters.
• I wish there was some kind of plugin system to add commands to doit. Requiring everyone to install matplotlib and networkx is not nice.
• what I really wish was some kind interactive visualization where you could hide/show nodes, select which type of nodes to show, cluster nodes, etc... but of course that's very ambitious.
• To start with I was thinking about just generate a text dot file and let graphviz handle the rendering. I am not familiar with networkx but it seems it works too :)

So, making it clear. This is very welcome but due to extra dependencies it cant come into doit core (it might have an "official" status but wont come in the main distribution package).
Something that generate a text file (like dot) that doesnt need extra packages would be accepted on core.

I will create another issue to discuss about a plugin system.

[schettino72]

issue #32 deals with a plugin system

[ankostis]

I think better leave the list command as it is now. And put graph as a separate command, it will probably grow to require its own parameters.

Myself originally had implemented it as a new graph command (i still have the code available) but came to pack it as list option because it really meddles with the workflow:

$ doit list
... 
doit list -g <a_task_from_above_list> 

And if its output is just a list of nodes, as you suggest, then it does make even more sense. I'm having git log as a guiding principle. Something like this: http://stackoverflow.com/a/9074343/548792

I wish there was some kind of plugin system to add commands to doit. Requiring everyone to install matplotlib and networkx is not nice.
...
To start with I was thinking about just generate a text dot file and let graphviz handle the rendering. I am not familiar with networkx but it seems it works too :)
...
Something that generate a text file (like dot) that doesnt need extra packages would be accepted on core.

• I fully agree for the plugn system.
• Matplotlib requirement is indeed too heavy and has to be factored out in a plugin.
• What i don't understand is the no-extra" package rule. The networkx is a pure-python 2 & 3 graph-library with very fast and lightweight data-structures, yet extremely powerful, perfectly fit to the tasks (reporting drawing, researching). I do not think any body will object in installing it.
  The only minor issue i see is that it is not tested on python-version prior to 2.7, but the documents still mention 2.6 and 3.2 as suitable versions. But is this such a showstopper for reinventing the graph-wheel?

[schettino72]

It is not just about adding another dependency. For a graph command I guess it makes sense to add the following options:

• drawer (one of networkx, dot, jpg, svg, json, etc)
• include file_dep, targets ?
• cluster files
• cluster subcommands
• i am sure gonna have more

these are just too much to go together with list command.

Also I have strict quality requirements for doit core, like 100% tests, etc. Leaving it outside gives it more freedom to evolve.

I added a basic plugin system 8091428
Check the test for an example of usage.

I suggest you not only move this code to a different command, it also should go into a different repository. I will invite you to join the pydoit group, but of course you are free to host it under your name if you prefer.

[schettino72]

@ankostis I sent an invitation but I am not 100% sure you have permissions to create a new repo... let me know. cheers

[ankostis]

Thank you, i accepted it.

[ankostis]

Ported functionality to new repo: https://github.com/pydoit/doit-graphx

[schettino72]

Great. I am closing this. At some point you should send an email to doit email group to announce this :)

[schettino72]

Also we need to feature it somewhere in the main doit docs...

[goretkin]

The homepage mentions

DAG Visualisation: create task's dependency-graph image using graphviz

This issue comes up in a search, so I thought it would be helpful to others if I ask here: how do I access this functionality?

[schettino72]

https://github.com/pydoit/doit-graph