added grammar parser for F object

[dbarrosop]

Implemented a grammar parser that turns a string into an F object. For instance:

• f = F(site="site1") is equivalent to f = parse("site == 'site1'")
• f = (F(site="site2") | F(role="www")) & F(my_var="comes_from_dev1.group_1") is equivalent to f = parse("(site == 'site2' OR role == 'www') AND my_var == 'comes_from_dev1.group_1'")

and long etc.

Look at the tests for more examples.

TODO:

1. Document
2. Get feedback
3. ~Fix CI error which is unrelated to the PR but just your usual python dependency hell~ (thanks to @ktbyers)

[dmulyalin]

Would be interested to hear what is the problem we trying to solve here?

Seems interesting solution, but struggling to come out with use case for it, that grammar seems quiet human readable as, IMHO, this:

1: f = parse("(site == 'site2' OR role == 'www') AND my_var == 'comes_from_dev1.group_1'")

easier to read/comprehand compared to

2: f = (F(site="site2") | F(role="www")) & F(my_var="comes_from_dev1.group_1")

however, what about this form: 3:

f = parse(
    [
        {"site": "site2", "my_var": "comes_from_dev1.group_1"}, 
        {"role": "www", "my_var": "comes_from_dev1.group_1"}
    ]
)

where, for instance, list of dictionaries express OR logic between items, but AND logic within each dictionary, that way host satisfying any of the dictionaries will pass the filter, equivalent to:

f = (
    (F(site="site2", my_var="comes_from_dev1.group_1") | 
    (F(role="www", my_var="comes_from_dev1.group_1"))
)

Option	Pros	Cons
1	Human readable	Hard to construct programmatically and need to learn grammar
2	Built into NR, already exists	Relatively hard to read and remember, hard to construct programmatically
3	Human readable and easy to construct/process programmatically	Need to remember implicit filtering logic, less flexible (?)

Also, to implement option 3, code might be:

def _filter_FO(nr, filter_data):
    """
    Function to filter hosts using Filter Object
    """
    ret = nr
    ret = ret.filter(F(**filter_data[0]))
    for item in filter_data[1:]:
        filtered_hosts = nr.filter(F(**item))
        ret.inventory.hosts.update(filtered_hosts.inventory.hosts)
    return ret

[dbarrosop]

This is for use in command lines (i.e. a —filter flag), apis, etc where you have no python.

Notice that this grammar parser doesn’t replace the F object, as a matter of fact, the grammar parser converts a string into it so without it it wouldn’t work :) The recommended way of filtering within python will still be using the F object but everybody seems to be inventing some way of describing F objects in text somehow for their command line tools so this is the response to that.

[dmulyalin]

Ic, makes more sense now, thanks for explanation.

[aedwardstx]

I love the idea of having a standard filter grammar parser. Here's my 2 cents:

• Consider making it so that if a raw hostname is supplied, it would be converted to "name==hostname”
• Consider making spaces optional between key and value e.g. “name==‘rtr1’”
• Consider making quotes optional for values e.g. “name==rtr1”
• NOT and != seem to overlap. Consider using key=value instead of == and keeping NOT
• Consider lower casing NOT/AND/OR or supporting both cases
• Is __endswith supported?

Some examples considering the above comments:

• rtr1
• name=rtr1
• name__startswith=rtr
• (group=routers or group=switches) and not platform=cisco_ios