Django Tastypie Advanced Filtering: How to do complex lookups with Q objects

You are on the right track. However, build_filters is supposed to transition resource lookup to an ORM lookup.

The default implementation splits the query keyword based on __ into key_bits, value pairs and then tries to find a mapping between the resource looked up and its ORM equivalent.

Your code is not supposed to apply the filter there only build it. Here is an improved and fixed version:

def build_filters(self, filters=None):    if filters is None:        filters = {}    orm_filters = super(BusinessResource, self).build_filters(filters)    if('query' in filters):        query = filters['query']        qset = (                Q(name__icontains=query) |                Q(description__icontains=query) |                Q(email__icontains=query)                )        orm_filters.update({'custom': qset})    return orm_filtersdef apply_filters(self, request, applicable_filters):    if 'custom' in applicable_filters:        custom = applicable_filters.pop('custom')    else:        custom = None    semi_filtered = super(BusinessResource, self).apply_filters(request, applicable_filters)    return semi_filtered.filter(custom) if custom else semi_filtered

Because you are using Q objects, the standard apply_filters method is not smart enough to apply your custom filter key (since there is none), however you can quickly override it and add a special filter called "custom". In doing so your build_filters can find an appropriate filter, construct what it means and pass it as custom to apply_filters which will simply apply it directly rather than trying to unpack its value from a dictionary as an item.

I solved this problem like so:

Class MyResource(ModelResource):  def __init__(self, *args, **kwargs):    super(MyResource, self).__init__(*args, **kwargs)    self.q_filters = []  def build_filters(self, filters=None):    orm_filters = super(MyResource, self).build_filters(filters)    q_filter_needed_1 = []    if "what_im_sending_from_client" in filters:      if filters["what_im_sending_from_client"] == "my-constraint":        q_filter_needed_1.append("something to filter")    if q_filter_needed_1:      a_new_q_object = Q()      for item in q_filter_needed:        a_new_q_object = a_new_q_object & Q(filtering_DB_field__icontains=item)      self.q_filters.append(a_new_q_object)  def apply_filters(self, request, applicable_filters):    filtered = super(MyResource, self).apply_filters(request, applicable_filters)    if self.q_filters:      for qf in self.q_filters:        filtered = filtered.filter(qf)      self.q_filters = []    return filtered

This method feels like a cleaner separation of concerns than the others that I've seen.

Taking the idea in astevanovic's answer and cleaning it up a bit, the following should work and is more succinct.

The main difference is that apply_filters is made more robust by using None as the key instead of custom (which could conflict with a column name).

def build_filters(self, filters=None):    if filters is None:        filters = {}    orm_filters = super(BusinessResource, self).build_filters(filters)    if 'query' in filters:        query = filters['query']        qset = (                Q(name__icontains=query) |                Q(description__icontains=query) |                Q(email__icontains=query)                )        orm_filters.update({None: qset}) # None is used as the key to specify that these are non-keyword filters    return orm_filtersdef apply_filters(self, request, applicable_filters):    return self.get_object_list(request).filter(*applicable_filters.pop(None, []), **applicable_filters)    # Taking the non-keyword filters out of applicable_filters (if any) and applying them as positional arguments to filter()