修改@ivanperelivskiy的回答:

@property
def _dict(self):
    ret = {}
    for field in self._meta.get_fields():
        if isinstance(field, ForeignObjectRel):
            # foreign objects might not have corresponding objects in the database.
            if hasattr(self, field.get_accessor_name()):
                ret[field.get_accessor_name()] = getattr(self, field.get_accessor_name())
            else:
                ret[field.get_accessor_name()] = None
        else:
            ret[field.attname] = getattr(self, field.attname)
    return ret

这里使用了django 1.10的公共方法get_fields。这使得代码更经得起未来的考验，但更重要的是还包括了外键和edititable =False的字段。

作为参考，这里是.fields的实现

@cached_property
def fields(self):
    """
    Returns a list of all forward fields on the model and its parents,
    excluding ManyToManyFields.

    Private API intended only to be used by Django itself; get_fields()
    combined with filtering of field properties is the public API for
    obtaining this field list.
    """
    # For legacy reasons, the fields property should only contain forward
    # fields that are not private or with a m2m cardinality. Therefore we
    # pass these three filters as filters to the generator.
    # The third lambda is a longwinded way of checking f.related_model - we don't
    # use that property directly because related_model is a cached property,
    # and all the models may not have been loaded yet; we don't want to cache
    # the string reference to the related_model.
    def is_not_an_m2m_field(f):
        return not (f.is_relation and f.many_to_many)

    def is_not_a_generic_relation(f):
        return not (f.is_relation and f.one_to_many)

    def is_not_a_generic_foreign_key(f):
        return not (
            f.is_relation and f.many_to_one and not (hasattr(f.remote_field, 'model') and f.remote_field.model)
        )

    return make_immutable_fields_list(
        "fields",
        (f for f in self._get_fields(reverse=False)
         if is_not_an_m2m_field(f) and is_not_a_generic_relation(f) and is_not_a_generic_foreign_key(f))
    )

我对@iperelivskiy的解决方案的看法是:在大规模的情况下，为每个__init__创建_initial字典是昂贵的，而且大多数时候是不必要的。我稍微改变了mixin，这样它只在你显式地告诉它这样做(通过调用instance.track_changes)时才记录更改:

from typing import KeysView, Optional
from django.forms import model_to_dict

class TrackChangesMixin:
    _snapshot: Optional[dict] = None

    def track_changes(self):
        self._snapshot = self.as_dict

    @property
    def diff(self) -> dict:
        if self._snapshot is None:
            raise ValueError("track_changes wasn't called, can't determine diff.")
        d1 = self._snapshot
        d2 = self.as_dict
        diffs = [(k, (v, d2[k])) for k, v in d1.items() if str(v) != str(d2[k])]
        return dict(diffs)

    @property
    def has_changed(self) -> bool:
        return bool(self.diff)

    @property
    def changed_fields(self) -> KeysView:
        return self.diff.keys()

    @property
    def as_dict(self) -> dict:
        return model_to_dict(self, fields=[field.name for field in self._meta.fields])

在我的解决方案是覆盖目标字段类的pre_save()方法之前，我有这种情况，它只在字段被更改时才会被调用 FileField有用 例子:

class PDFField(FileField):
    def pre_save(self, model_instance, add):
        # do some operations on your file 
        # if and only if you have changed the filefield

劣势: 如果你想做任何(post_save)操作，比如在某些作业中使用创建的对象(如果某些字段已更改)，则无用

有时我想检查多个共享这些字段的模型上相同特定字段的更改，因此我定义了这些字段的列表并使用一个信号。在这种情况下，只有当某些东西发生了变化，或者条目是新的时，地理编码才会寻址:

from django.db.models.signals import pre_save
from django.dispatch import receiver

@receiver(pre_save, sender=SomeUserProfileModel)
@receiver(pre_save, sender=SomePlaceModel)
@receiver(pre_save, sender=SomeOrganizationModel)
@receiver(pre_save, sender=SomeContactInfoModel)
def geocode_address(sender, instance, *args, **kwargs):

    input_fields = ['address_line', 'address_line_2', 'city', 'state', 'postal_code', 'country']

    try:
        orig = sender.objects.get(id=instance.id)
        if orig:
            changes = 0
            for field in input_fields:
                if not (getattr(instance, field)) == (getattr(orig, field)):
                    changes += 1
            if changes > 0:
                # do something here because at least one field changed...
                my_geocoder_function(instance)
    except:
        # do something here because there is no original, or pass.
        my_geocoder_function(instance)

只编写一次并使用“@receiver”附加当然胜过重写多个模型保存方法，但也许其他人有更好的想法。

最好的方法是使用pre_save信号。在09年这个问题被提出并回答的时候，可能还没有这个选项，但今天看到这个问题的人应该这样做:

@receiver(pre_save, sender=MyModel)
def do_something_if_changed(sender, instance, **kwargs):
    try:
        obj = sender.objects.get(pk=instance.pk)
    except sender.DoesNotExist:
        pass # Object is new, so field hasn't technically changed, but you may want to do something else here.
    else:
        if not obj.some_field == instance.some_field: # Field has changed
            # do something

@ivanlivski的mixin很棒。

我把它扩展到

确保它适用于Decimal字段。 公开属性以简化使用

更新后的代码可在这里: https://github.com/sknutsonsf/python-contrib/blob/master/src/django/utils/ModelDiffMixin.py

为了帮助刚接触Python或Django的人，我将给出一个更完整的示例。 这种特殊用法是从数据提供者获取一个文件，并确保数据库中的记录反映该文件。

我的模型对象:

class Station(ModelDiffMixin.ModelDiffMixin, models.Model):
    station_name = models.CharField(max_length=200)
    nearby_city = models.CharField(max_length=200)

    precipitation = models.DecimalField(max_digits=5, decimal_places=2)
    # <list of many other fields>

   def is_float_changed (self,v1, v2):
        ''' Compare two floating values to just two digit precision
        Override Default precision is 5 digits
        '''
        return abs (round (v1 - v2, 2)) > 0.01

加载文件的类有这些方法:

class UpdateWeather (object)
    # other methods omitted

    def update_stations (self, filename):
        # read all existing data 
        all_stations = models.Station.objects.all()
        self._existing_stations = {}

        # insert into a collection for referencing while we check if data exists
        for stn in all_stations.iterator():
            self._existing_stations[stn.id] = stn

        # read the file. result is array of objects in known column order
        data = read_tabbed_file(filename)

        # iterate rows from file and insert or update where needed
        for rownum in range(sh.nrows):
            self._update_row(sh.row(rownum));

        # now anything remaining in the collection is no longer active
        # since it was not found in the newest file
        # for now, delete that record
        # there should never be any of these if the file was created properly
        for stn in self._existing_stations.values():
            stn.delete()
            self._num_deleted = self._num_deleted+1


    def _update_row (self, rowdata):
        stnid = int(rowdata[0].value) 
        name = rowdata[1].value.strip()

        # skip the blank names where data source has ids with no data today
        if len(name) < 1:
            return

        # fetch rest of fields and do sanity test
        nearby_city = rowdata[2].value.strip()
        precip = rowdata[3].value

        if stnid in self._existing_stations:
            stn = self._existing_stations[stnid]
            del self._existing_stations[stnid]
            is_update = True;
        else:
            stn = models.Station()
            is_update = False;

        # object is new or old, don't care here            
        stn.id = stnid
        stn.station_name = name;
        stn.nearby_city = nearby_city
        stn.precipitation = precip

        # many other fields updated from the file 

        if is_update == True:

            # we use a model mixin to simplify detection of changes
            # at the cost of extra memory to store the objects            
            if stn.has_changed == True:
                self._num_updated = self._num_updated + 1;
                stn.save();
        else:
            self._num_created = self._num_created + 1;
            stn.save()