So we do lose set semantics here, can you add a test with multiple hooks that have the same additional dependencies and verify that there are not duplicates?

Well, the current set implementation is already accepting the same dependency listed at different versions. We could keep the same behaviour by handling duplicates manually:

for dep in hook.get('additional_dependencies', []):
    if dep not in dep_dict[hook['language']][hook['language_version']]:
        dep_dict[hook['language']][hook['language_version']].append(dep)

An other option would be to extract the name and the version for each dependency (needs to be implemented for each language though) and then either:

• keep only the first version of each package
• issue a warning to the user
• raise an exception

and then implement tests accordingly, what would you prefer?

The former is fine, I don't want to add language-specific parsing of this stuff if avoidable.

Heh, if we could only target python3.6 (or modern pypy), the set would have insertion order

A warning should not be raised here.

This is a perfectly valid configuration:

    hooks:
    -   id: puppet-validate
        language_version: '2.1.5'
        additional_dependencies: ['puppet:4.5.3']
    -   id: epp-validate
        language_version: '2.1.5'
        additional_dependencies: ['puppet:4.5.3']

Maybe we can keep the code change smaller by making something like this:

class _UniqueList(list):
    def __init__(self):
        self._set = set()

    def update(self, obj):
        for item in obj:
            if item not in self._set:
                self.append(item)

And then the diff is just

-        dep_dict = defaultdict(lambda: defaultdict(set))
+        dep_dict = defaultdict(lambda: defaultdict(_UniqueList))

👍

Let me know if you want me to squash commits

Up to you!