[capink]

Originally Posted by chaley

It doesn't fail for me.

Is the 'P' in 'Petra' a non-Latin unicode character? I can see something like this happening if it is a unicode composed character.

FWIW: there is some strange (to me) code in your function:

def evaluate(self, formatter, kwargs, mi, locals, val, col_name, user_cat_prefix):
    new_val = ''
    if hasattr(mi, '_proxy_metadata'):
        # Below should be mi.user_categories
        all_cats = mi._proxy_metadata.user_categories
        cats = {k:v for k,v in all_cats.items() if k.startswith(user_cat_prefix)}
        SEP = mi.metadata_for_field(col_name)['is_multiple'].get('list_to_ui', '')
        new_val = set()
        if SEP:
            val_ = val.split(SEP)
        else:
            val_ = [val]  
        for user_cat, v in cats.items():
            repl = user_cat.lstrip(user_cat_prefix)
            for user_cat_item, src_cat in v:
                if src_cat == col_name:
                    # Why does the next line have the slice?
                    # I think it should be 
                    # for item in val_:
                    for item in val_[:]:
                        if item == user_cat_item:
                            new_val.add(repl)
                            # Why are you removing the item from the list?
                            # Sometimes that will mess up the for loop. Or
                            # at least it could in python 2.
                            val_.remove(item)
                        else:
                            new_val.add(item)
        if new_val:
            return ', '.join(list(new_val))
    return val

I wrote this a long time ago, so I don't remember the details. The slice was intentional to create a new list (a substitute for copy.copy()), because I needed to preserve the original one (I don't remember the details now).

Edit: Looking at the code now, I guess the slicing might have been an attempt to improve performance of the template function, by not having to iterate over items that have been already processed.

The mi._proxy_metadata.user_categories was the way I saw it done in one of the builtin template functions. But since you recommend mi.user_categories(), I will be using that instead.