• This should probably document that the items must be hashable.
• I'm not sure there's a good enough usecase for this. What about {{ set(x)|sort }}?
• It'd probably make sense to implement this as separate filter.

Re 1: Indeed, I didn't check the details.

Re 2: I haven't seen this anywhere in the documentation and I am new to Jinja enough not to know which parts of it are common with Python. Too bad it doesn't work.

Traceback (most recent call last):
  File "foo.py", line 526, in <module>
    timeoffs = timeoffs,
  File "/usr/lib/python2.7/dist-packages/jinja2/environment.py", line 894, in render
    return self.environment.handle_exception(exc_info, True)
  File "templates/weekly_for_hq.html.j2", line 83, in top-level template code
    {# Staff #}<td>{{ set(engineers)|sort|join(', ') }}</td>
jinja2.exceptions.UndefinedError: 'set' is undefined

Re 3: I thought about it. However, I prefered sorted output anyway so I came up with something simmilar to sort -u.

Re 2: I'm not so sure anymore if that builtin is actually available in Jinja. I think it's not.
Re 3: On the CLI there's also uniq available.

FWIW you can add your own filters and globals to Jinja, and since Python doesn't have a comparable function, I am not sure if this might not be out of scope -- even if only slightly. You're the first one to request this, so I'd like to just keep this open and see how many people +1 this.

I'd agree that a built-in way to eliminate duplicates from a sequence would be desirable. In fact, I ended up implementing a unique filter myself for a couple of projects, and already asked to have it added to upstream jinja2, 4 years ago as part of #66, see snoack/jinja2@2d9a896. But I think it should be a separate filter for larger flexibility. So one can eliminate duplicates from a sequence while preserving order.

I have just rebased and improved my old patch, and submitted a new pull request, see #469.

I'm not against adding a filter like unique or set, but like @untitaker I'd rather see it as a separate filter (as implemented in #469) rather than as a param on sort.