this should solve issues presented in #7495, without a need for a workaround

Would it be possible to check the types of the arguments instead of the try/except?

I can check for dicts, but other types can be passed that are not
'hashable', the try/except covers all possible ones.

But considering that most of this should be comming from yaml, it should be
limited to dicts.

Brian Coca
Stultorum infinitus est numerus
0110000101110010011001010110111000100111011101000010000001111001011011110111010100100000011100110110110101100001011100100111010000100001
Pedo mellon a minno

actually, after thinking for 10s:

if isinstance(a, collections.Hashable) : ...
​

thanks, I had changed the rest, I'm not sure why I missed this one

Actually @mpdehaan pointed out that, by turning inputs into sets, we are removing duplicates in them too. Hence, the fallback operations are all wrong here.

Example with hashable items:

a,b = [x,x,y,y], [x,z,z]
unique(a) # set([x,y])
intersect(a, b) # set([x])
difference(a, b) # set([y])
symmetric_difference(a, b) # set([y,z])
union(a, b) # set([x,y,z])

Example with non-hashable items:

a,b = [x,x,y,y], [x,z,z]
unique(a) # set([x,x,y,y]) - Wrong: the not in c check is useless because c is populated only
                             after finishing the whole filter operation
intersect(a, b) # set([x,x]) - Wrong: duplicates in a are not removed
difference(a, b) # set([y,y]) - Wrong: duplicates in a are not removed
symmetric_difference(a, b) # set([y,y,z,z]) - Wrong: duplicates not removed from neither a nor b
union(a, b) # set([x,y,z]) # set([x,x,y,y,x,z,z]) - Wrong: duplicates not removed from neither a nor b

Also, the exact same result (than the set operations) is impossible to achieve because sets are implicitly unordered (meaning that you can get varying orders when iterating it) whereas lists are ordered by definition. Anyway, I guess this is an acceptable trade-off...

I'll resubmit tonight, also making changes to try to coerce always
returning a list/set when used in templated fields (with_items:).​

This even fixes an issue when using filters in with_items (e.g. #8055)

This patch fixes #8141 for me. +1

+1 as we keep seeing people bitten by the use of non hashable objects with the set theory filters

Merged, thanks!