Trim trailing whitespace

[jayvdb]

Similar to #36 , and the trailing blank line fixup, ..

When importing content from other providers, their trailing blank lines shouldnt be propagated into moban managed repositories.

e.g. github/gitignore#2728 and moremoban/pypi-mobans#22 (comment)

If trailing whitespace needs to be preserved, we can use jinja {% raw %} to keep it.

[chfw]

Hmmm good point. Can have a look.

[jayvdb]

This could be a simple template plugin, like copy.

When it needs to be added to a templated file, a pipeline could be used. i.e. the mobanfile can have an extra target for intermediate file, like how gmake works.

e.g. .jj2 -> .dirty.txt -> .txt

and the intermediate file could be removed by a delete rule (#167)

Or we might allow the same target to be listed twice, and then use two template engines consecutively (jinja, then whitespace-cleaner).

[chfw]

We can introduce filter plugin but my worry is that current implementation does not respect(or guarantee) appearance order. So all is collected and aggregated to a dictionary: template type: its template targets. And then this dictionary is iterated through its corresponding template engine.

[jayvdb]

I think that problem needs to be solved, and solved using a 'gmake' like algorithm.

We should be able to find an existing 'solver' algorithm, maybe by looking in other tools like https://github.com/brushtechnology/fabricate , or https://github.com/mesonbuild/meson (but the solver might be in https://github.com/ninja-build/ninja )

SAT solvers are another approach, but not so well supported in Python and probably overkill . https://research.swtch.com/version-sat

[jayvdb]

We could have a very basic solution by adding a 'plugin order' config value to the yaml, so the user can specify

moban_plugin_order:
  - jinja
  - trim-whitespace
  - etc

[chfw]

I think ordereddict should be able to help us.

[chfw]

one more missing functionality: moban expects all templates to have existed before execution. In this case, one output of template engine become the input of another, this will cause exception for current moban.

[chfw]

And to truly respect the template file based sequence order( rather than template type based sequence order), two alternatives can be considered:

a. ignore moban engine optimization where if found that a data config or a template file is used across many templating actions, the file will be re-used across these actions.

b. provide higher level of abstraction os that render_to_files does not get invoked immediately but only generate an execution sequence, and such a sequence is given to a batch runner which does the action no matter which templating engine is involved. Then, we can consider include a in the batch runner so that render_to_files will drop the optimization and batch runner will try to re-implement the optimization.