New check: Use single-statement dict initialization

[nickdrozd]

Here's an ugly pattern, commonly found in eg config files:

config = {}
config['dir'] = 'bin'
config['user'] = 'me'
config['workers'] = 5

An empty dict is initialized and then mutated several times. It would
be better to use declarative literal syntax:

config = {
    'dir': 'bin',
    'user': 'me',
    'workers': 5,
}

No mutation! No repetition!

I'm imagining a check that would look for an empty dict initialization
and then see if any following lines are bare dict assignments. It
could get tricky though, as this pattern horrible pattern is sometimes
recursive:

config = {}
config['dir'] = 'bin'
config['user'] = 'me'
config['workers'] = 5
config['options'] = {}
config['options']['debug'] = False
config['options']['verbose'] = True

Yuck!

And I know, a check like this gets released, CIs start showing failing
builds, and everyone thinks the sky is falling. Maybe make it
optional? Or a warning or something? Anything to discourage this
pattern.

[PCManticore]

Sounds like a good improvement @nickdrozd This can be an optional check for now, in the form of an extension pylint.checkers.extensions, which needs to be explicitly enabled in the configuration file. Maybe we can enable it by default in pylint 3.0.

[nickdrozd]

@PCManticore I'd like to work on this one, but I'm not sure exactly where to start. What kind of node would this check attach to? One approach would be to look for assign statements with {} on the right side, then collect all immediately following assign statements where the right side indexes into the variable assigned to by the first statement. Does that sound right? Are there any existing checks that do something similar?

[PCManticore]

@nickdrozd That approach sounds right to me. You should find some similar checks in checkers.refactoring, consider-using-get and simplifiable-if-statement come to mind.

[nickdrozd]

I changed the title of this issue to accommodate #4365