Refactor features

[brettviren]

I found an old thread initiated by Sebastien on the waf list. It shows a possibly better way to handle features

https://groups.google.com/d/msg/waf-users/eS4NGebR1Zc/iDY3zSFtWewJ

It points out the abstract_process example in the waf source.

worch features were originally something closer to this but my lack of understanding and some problems with dependency resolution led me to the current setup. Using waf features instead of made up worch ones and using explicit marker files for each step seems to be more organized way to do things..

[brettviren]

I've started a "feature-cleanup" branch for this.

[brettviren]

My thinking on this is:

• rewrite existing features as real waf features
• add a "tools" configuration keyword that will be interpreted to load outside tools to gain new features
• rewrite existing feature taskgens as lower level tasks so they can be more easily shared by different features
• rewrite the mess that is pfi.py

[brettviren]

The current feature/requirements.py constructs a lot of directory convention. This is confusing and requires the pfi mess I plan to remove both but this will mean configurations need to be re-tweaked.

[brettviren]

This has been merged into master. A few changes worth noting:

• worch features are now waf features
• features can be defined in waf tools and loaded based on the "tools" configuration item
• tool code can specify default values for any configuration parameter
• every rule that ends up invoking the worch exec_command will have the output placed in {out}/logs/ instead of CWD
• all waf tasks should be defined through the step() method augmented onto the task gen. This will append a standard "control" file as output. This file is named by convention and a node for it can be got from control_node() augmented method in order for features to insert dependencies without needing to know details about what files a step otherwise might produce.