Djaq 0.1.3

A string-based Django query language

Djaq - pronounced “Jack” - provides an instant remote API to your Django models data with a powerful query language. No server-side code beyond two lines of config is required. You don’t write backend classes, serializers or any other code to be able to immediately get whatever data you want to the client. And it is blazing fast.

Djaq queries are strings. A query string for our example dataset might look like this:

(b.name as title, b.publisher.name as publisher) Book b

This retrieves a list of book titles with book publisher. But you can formulate far more sophisticated queries, see below. You can send Djaq queries from any language, Java, Javascript, golang, etc. to a Django application and get results as JSON. In contrast to REST frameworks, like TastyPie or Django Rest Framework (DRF), you have natural access to the Django ORM from the client.

Djaq is a good fit if you want:

• Microservice communication where some services don’t have access to the Django ORM or are not implemented with Python

• Fast local UI development

• Fast development of Proof of Concepts

Djaq sits on top of the Django ORM. It can happily be used alongside QuerySets and sometimes calling a Djaq query even locally might be preferable to constructing a Queryset, although Djaq is not a replacement for QuerySets.

Features you might appreciate:

• Immediate gratification with zero or minimal server-side code. Because there is minimal setup, there is minimal wasted effort if you later move to another framework, like GraphQL or DRF. But getting started calling your API is much faster than those frameworks.

• A natural syntax that lets you compose queries using Python-like expressions. The query format and syntax is designed to be written by hand quickly. Readability is a key goal.

• Complex expressions let you push computation down to the database layer from the client easily.

• Fast cursor semantics and explicit retrieval. It only gets data you asked for.

• Obvious performance behaviour. It will trigger a query in one obvious way through one of the generator methods: .dict(), .tuples(), .json().

• A ready-to-go CRUD API that is easy to use. You can send requests to have an arbitrary number of Create, Read, Write, Delete operations done in a single request.

• Customisable behaviour using your own functions and data validators.

• A handy user interface for trying out queries on your data models.

Djaq provides whitelisting of apps and models you want to expose. It also provides a simple permissions scheme via settings.

Note that Djaq is still in an early phase of development. No warranties about reliability, security or that it will work exactly as described.

Djaq UI

Quickstart and Installation

You need Python 3.6 or higher and Django 2.1 or higher.

Install:

pip install Djaq

The bleeding edge experience:

pip install https://github.com/paul-wolf/djaq/archive/master.zip

Use:

from djaq.query import DjangoQuery as DQ

print(list(DQ("(b.name as title, b.publisher.name as publisher) Book b").dicts()))

[{'title': 'Name grow along.', 'publisher': 'Long, Lewis and Wright'}, {'title': 'We pay single record.', 'publisher': 'Long, Lewis and Wright'}, {'title': 'Natural develop available manager.', 'publisher': 'Long, Lewis and Wright'}, {'title': 'Fight task international.', 'publisher': 'Long, Lewis and Wright'}, {'title': 'Discover floor phone.', 'publisher': 'Long, Lewis and Wright'}]

Providing an API

We’ll assume below you are installing the Djaq UI. This is not required to provide an API but is very useful to try things out.

Install the API and UI in settings:

INSTALLED_APPS = (
    ...
    djaq.djaq_api,
    djaq.djaq_ui,
)

Configure urls in urls.py:

urlpatterns = [
    ...
    path("dquery/", include("djaq.djaq_ui.urls")),`
    path("djaq/", include("djaq.djaq_api.urls")),`
]

You are done. You can start sending requests to:

/djaq/api/request/

The UI will be available at:

/dquery

Note the UI will send requests to the API endpoint so will not work without that being configured. You send a request in this form to the api endpoint:

{
 "queries": [
  {
   "q": "(b.id,b.name,b.pages,b.price,b.rating,b.publisher,b.alt_publisher,b.pubdate,b.in_print,) books.Book b",
   "context": {},
   "limit": "100",
   "offset": "0"
  }
 ]
}

The UI will create this JSON for you if you want to avoid typing it.

You can also create objects, update them and delete them:

{
   "queries": [
    {
      "q": "(b.id,b.name,b.pages,b.price,b.rating,b.publisher,b.alt_publisher,b.pubdate,b.in_print,) books.Book b",
      "context": {},
      "limit": "100",
      "offset": "0"
    }
   ],
  "creates":[{
     "_model":"Book"
     "name":"my new book",
     }],
  "updates":[{
     "_model":"Book"
     "_pk": 37,
     "name":"my new title",
     }],
  "deletes":[{
     "_model":"Book"
       "_pk": 37,
     }]
}

You can send multiple queries, creates, updates, deletes operations in a single request.

Settings

The API and UI will use the following settings:

• DJAQ_WHITELIST: a list of apps/models that the user is permitted to include in queries.

• DJAQ_PERMISSIONS: permissions required for staff and superuser.

• DJAQ_VALIDATOR: if using the remote API, you can specify a validator class to handle all requests. The value assigned must be a class derived from djaq.query.ContextValidator. The request object is always added to the context by default. You can examine this in the validator to make decisions like forbidding access to some users, etc.

In the following example, we allow the models from ‘books’ to be exposed as well as the User model. We also require the caller to be both a staff member and superuser:

DJAQ_WHITELIST = {
    "django.contrib.auth": ["User"],
    "books": [
        "Profile",
        "Author",
        "Consortium",
        "Publisher",
        "Book_authors",
        "Book",
        "Store_books",
        "Store",
    ],
}
DJAQ_PERMISSIONS = {
    "creates": True,
    "updates": True,
    "deletes": True,
    "staff": True,
    "superuser": True,
}

If we want to allow all models for an app, we can leave away the list of models. This will have the same effect as the setting above.

DJAQ_WHITELIST = {
    "django.contrib.auth": ["User"],
    "books": [],
}

For permissions, you can optionally require any requesting user to be staff and/or superuser. And you can deny or allow update operations. If you do not provide explicit permissions for update operations, the API will respond with 401 if one of those operations is attempted.

Custom API

You can write your own custom API endpoint. Here is what a view function for your data layer might look like with Djaq:

@login_required
def djaq_view(request):
    data = json.loads(request.body.decode("utf-8"))
    query_string = data.get("q")
    offset = int(data.get("offset", 0))
    limit = int(data.get("limit", 0))
    context = data.get("context", {})
    return JsonResponse(
        {
           "result": list(
               DQ(query_string)
               .context(context)
               .limit(limit)
               .offset(offset)
               .dicts()
           )
        }
    )

You can now query any models in your entire Django deployment remotely, provided the authentication underlying the login_required is satisfied. This is a good solution if your endpoint is only available to trusted clients who hold a valid authentication token or to clients without authentication who are in your own network and over which you have complete control. It is a bad solution on its own for any public access since it exposes Django framework models, like users, permissions, etc.

Most likely you want to control access in two ways:

• Allow access to only some apps/models

• Allow access to only some rows in each table and possibly only some fields.

For controlling access to models, use the whitelist parameter in constructing the DjangoQuery:

DQ(query_string, whitelist={"books": ["Book", "Publisher",],})
  .context(context)
  .limit(limit)
  .offset(offset)
  .dicts()

This restricts access to only the book app models, Book and Publish.

You probably need a couple more things if you want to expose this to a browser. But this gives an idea of what you can do. The caller now has access to any authorised model resource. Serialisation is all taken care of. Djaq comes already with a view similar to the above. You can just start calling and retrieving any data you wish. It’s an instant API to your application provided you trust the client or have sufficient access control in place.

Difference between Djaq and Other Frameworks

The core of Djaq does not actually have anything specifically to do with remote requests. It is primarily a query language for Django models. You can just as easily use it within another remote API framework.

The default remote API for Djaq is not a REST framework. It does use JSON for encoding data and POST to send requests. But it does not adhere to the prescribed REST verbs. It comes with a very thin wrapper for remote HTTP(S) requests that is a simple Django view function. It would be trivial to write your own or use some REST framework to provide this functionality. Mainly, it provides a way to formulate queries that are highly expressive, compact and readable.

There is only one endpoint for Djaq on the backend.

Requests for queries, creates, updates, deletes are always POSTed.

Most importantly, the client decides what information to request using a query language that is much more powerful than what is available from other REST frameworks and GraphQL.

Conversely, REST frameworks and GraphQL are more useful than Djaq in providing server-side business rule implementation.

Limitations

Compared to other frameworks like GraphQL and DRF, you can’t easily implement complex business rules on the server. This might be a deal breaker for your application.

Djaq, without any configuration, provides access to all your model data. That is usually not what you want. For instance, you would not want to expose all user data, session data, or many other kinds of data to even authenticated clients. It is trivial to prevent access to data on an app or a model class level. But this might be too coarse-grained for your application.

Djaq only supports Postgresql at this time.

Performance

You will probably experience Djaq calls as blazing fast compared to other remote frameworks. This is just because not much happens inbetween. Once the query is parsed, it is about as fast as you will ever get unless you do something fancy in a validator. The simplest possible serialization is used by default.

Once the query is parsed, it is about the same overhead as calling this:

conn = connections['default']
cursor = conn.cursor()
self.cursor = self.connection.cursor()
self.cursor.execute(sql)

Parsing is pretty fast and might be a negligible factor if you are parsing during a remote call as part of a view function.

But if you want to iterate over, say, a dictionary of variables locally, you’ll want to parse once:

dq = DQ("(b.name) Book{ilike(b.name, '$(namestart)')} b")
dq.parse()
for vars in var_list:
    results = list(dq.context(vars).tuples())
    <do something with results>

Note that each call of context() causes the cursor to execute again when tuples() is iterated.

Query usage guide

Throughout, we use models somewhat like those from Django’s bookshop example:

from django.db import models

class Author(models.Model):
    name = models.CharField(max_length=100)
    age = models.IntegerField()

class Publisher(models.Model):
    name = models.CharField(max_length=300)

class Book(models.Model):
    name = models.CharField(max_length=300)
    pages = models.IntegerField()
    price = models.DecimalField(max_digits=10, decimal_places=2)
    rating = models.FloatField()
    authors = models.ManyToManyField(Author)
    publisher = models.ForeignKey(Publisher, on_delete=models.CASCADE)
    pubdate = models.DateField()

class Store(models.Model):
    name = models.CharField(max_length=300)
    books = models.ManyToManyField(Book)

These examples use auto-generated titles and names and we have a slightly more complicated set of models than shown above.

Let’s get book title (name), price, discounted price, amount of discount and publisher name wherever the price is over 50.

result = \
  list(DQ("""(b.name,
       b.price as price,
       0.2 as discount,
       b.price * 0.2 as discount_price,
       b.price - (b.price*0.2) as diff,
       Publisher.name
      ) Book{b.price > 50} b""").dicts())

result now contains a list of dicts each of which is a row in the result set. One example:

[{'b_name': 'Address such conference.',
  'price': Decimal('99.01'),
  'discount': Decimal('0.2'),
  'discount_price': Decimal('19.802'),
  'diff': Decimal('79.208'),
  'publisher_name': 'Arnold Inc'}]

Here is the structure of the syntax:

(<field_exp1>, ...) <ModelName>{<filter_expression>} <alias> order by (<field_exp1>, ...)

Whitespace does not matter too much. You could put things on separate lines:

(
   b.name, b.price,
   Publisher.name
)
Book{p.price > 50} b

Always start with column expressions you want to return in parens:

(b.name, b.price, Publisher.name)

These expressions can be Django Model fields or arithmetic expressions or any expression supported by functions of your underlying database that are also whitelisted by Djaq. Postgresql has thousands of functions. About 350 of those are available in Djaq.

Columns are automatically given names. But you can give them your own name:

(b.name as title, b.price as price, Publisher.name as publisher)

Next is the model alias declaration:

Book b

or if we want to filter and get only books over 50 in price:

Book{b.price > 50} b

Book is the Django Model name. b is an alias we can use as an abbreviation in the filter or returned column expressions. We put the filter in curly braces, {}, between the model name and alias. Use Python syntax to express the filter. Also use Python syntax to express the data to return. You don’t have access to the Python Standard Library. This is basically the intersection of SQL and Python:

The following filter:

{b.price > 50 and ilike(Publisher.name, 'A%')}

will be translated to SQL:

b.price > 50 AND publisher.name ILIKE 'A%'

The expressions are fully parsed so they are not subject to SQL injection. Trying to do so will cause an exception.

You might notice in the above examples, Publisher does not use an alias. If you wanted an alias for Publisher, you could use a more complicated syntax:

(b.name, b.price) Book b
-> (p.name) Publisher.name p

Notice, we use the -> symbol to add another aliased relationship. This is one of three options: ->, <-, <> that indicate you want to explicitly join via an SQL LEFT, RIGHT or INNER join respectively. But you don’t need to do this. LEFT joins will always be implicit. We did not even need to refer to the Publisher model directly. We could have done this:

(b.name, b.price, b.publisher.name as publisher)
Book{p.price > 50} b

Our example model also has an owner model called “Consortium” that is the owner of the publisher:

In [16]: print(list(DQ("(b.name, b.price, b.publisher.name, b.publisher.owner.name) Book b").limit(1).dicts()))
Out[16]: [{'b_name': 'Range total author impact.', 'b_price': Decimal('12.00'), 'b_publisher_name': 'Wright, Taylor and Fitzpatrick', 'b_publisher_owner_name': 'Publishers Group'}]

To recap, there are three alternative patterns to follow to get the publisher name in the result set:

In [13]: print(list(DQ("(b.name, b.price) Book b -> (p.name)Publisher p").limit(1).dicts()))

In [14]: print(list(DQ("(b.name, b.price, Publisher.name) Book b").limit(1).dicts()))

In [15]: print(list(DQ("(b.name, b.price, b.publisher.name) Book b").limit(1).dicts()))

Note that the above will each produce slightly different auto-generated output names unless you provide your own aliases.

Signal that you want to summarise results using an aggregate function:

list(DQ("(b.publisher.name as publisher, count(b.id) as book_count) Book b").dicts())

[
    {
        "publisher": "Martinez, Clark and Banks",
        "book_count": 6
    },
    {
        "publisher": "Fischer-Casey",
        "book_count": 9
    },
    etc.
]

Order by name:

(b.name, b.price, b.publisher.name as publisher)
Book{p.price > 50} b
order by (b.name)

The parentheses around the order by expression are required.

Get average, minimum and maximum prices:

list(DQ("(avg(b.price) as average, min(b.price) as minimum, max(b.price) as maximum) Book b).dicts())
[
   {
      "average": "18.5287169247794985",
      "minimum": "3.00",
      "maximum": "99.01"
   }
]

Count all books:

list(DQ("(count(b.id)) Book b").dicts())

[
    {
        "countb_id": 149999
    }
]

You can qualify model names with the app name or registered app path:

(b.name, b.publisher.name) books.Book b

You’ll need this if you have models from different apps with the same name.

To pass parameters, use variables in your query, like '$(myvar)':

In [30]: oldest = '2018-12-20'
    ...: list(DQ("(b.name, b.pubdate) Book{b.pubdate >= '$(oldest)'} b").context({"oldest": oldest}).limit(5).tuples())
Out[30]:
[('Available exactly blood.', datetime.date(2018, 12, 20)),
 ('Indicate Congress none always.', datetime.date(2018, 12, 24)),
 ('Old beautiful three program.', datetime.date(2018, 12, 25)),
 ('Oil onto mission.', datetime.date(2018, 12, 21)),
 ('Key same effect me.', datetime.date(2018, 12, 23))]

Notice that the variable holder, $(), must be in single quotes.

Query UI

You can optionally install a query user interface to try out queries on your own data set:

• After installing djaq, add djaq.djaq_ui to INSTALLED_APPS

• Add path("dquery/", include("djaq.djaq_ui.urls")), to urlpattenrs in the sitesurls.py`

Navigate to `/dquery/’ in your app and you should be able to try out queries.

• Send: call the API with the query

• JSON: show the json that will be sent as the request data

• SQL: show how the request will be sent to the database as sql

• Schema: render the schema that describe the available fields

• Whitelist: show the active whitelist. You can use this to generate a whilelist and edit it as required.

There is a combo dropdown control, apps. Select the Django app. Models for the selected app are listed below. If you click once on a model, the result field will show the schema for that model. If you double-click the model, it generates a query for you for all fields in that model. Once you do that, just press “Send” to see the results.

If the query pane has the focus, you can press shift-return to send the query request to the server.

Functions

If a function is not defined by DjangoQuery, then the function name is checked with a whitelist of functions. There are approximately 350 functions available.

A user can define new functions at any time by adding to the custom functions. Here’s an example of adding a regex matching function:

DjangoQuery.functions["REGEX"] = "{} ~ {}"

Now find all book names starting with ‘B’:

DQ("(b.name) Book{regex(b.name, 'B.*')} b")

We always want to use upper case for the function name when defining the function. Usage of a function is then case-insensitive. You may wish to make sure you are not over-writing existing functions. “REGEX” already exists, for instance.

You can also provide a callable to DjangoQuery.functions. The callable needs to take two arguments: the function name and a list of positional parameters and it must return SQL as a string that can either represent a column expression or some value expression from the underlying backend.

In the following:

DQ("(b.name) Book{like(upper(b.name), upper('$(name_search)'))} b")

like() is a Djaq-defined function that is converted to field LIKE string. Whereas upper() is sent to the underlying database because it’s a common SQL function. Any function can be created or existing functions mutated by updating the DjangoQuery.functions dict where the key is the upper case function name and the value is a template string with {} placeholders. Arguments are positionally interpolated.

Above, we provided this example:

DQ("""(
   sum(iif(b.rating < 5, b.rating, 0)) as below_5,
   sum(iif(b.rating >= 5, b.rating, 0)) as above_5
) Book b""")

We can simplify further by creating a new function. The IIF function is defined like this:

"CASE WHEN {} THEN {} ELSE {} END"

We can create a SUMIF function like this:

DjangoQuery.functions['SUMIF'] = "SUM(CASE WHEN {} THEN {} ELSE {} END)"

Now we can rewrite the above like this:

DQ("""(
    sumif(b.rating < 5, b.rating, 0) as below_5,
    sumif(b.rating >= 5, b.rating, 0) as above_5
    ) Book b""")

Here’s an example providing a function:

def concat(funcname, args):
    """Return args spliced by sql concat operator."""
    return " || ".join(args)

DjangoQuery.functions['CONCAT'] = concat

Parameters and Validator

We call the Django connection cursor approximately like this:

from django.db import connections
cursor = connections['default']
cursor.execute(sql, context_dict)

When we execute the resulting SQL query, named parameters are used. You must name your parameters. Positional parameters are not passed:

oldest = '2000-01-01'
DQ("(b.id) Book{b.pub_date >= '$(oldest)'} b").context({"oldest": oldest}).tuples()

Notice that any parameterised value must be represented in the query expression in single quotes:

'$(myparam)'

Therefore, when you add subqueries, their parameters have to be supplied at the same time.

Note what is happening here:

name_search = 'Bar.*'
DQ("(b.id) Book{regex(b.name, '%(name_search)')} b").context(locals()).tuples()

To get all books starting with ‘Bar’. Or:

DQ("(b.name) Book{like(upper(b.name), upper('$(name_search)'))} b").context(request.POST)

Provided that request.POST has a name_search key/value.

You can provide a validation class that will return context variables. The default class used is called ContextValidator(). You can override this to provide a validator that raises exceptions if data is not valid or mutates the context data, like coercing types from str to int:

class MyContextValidator(ContextValidator):
    def get(self, key, value):
        if key == 'order_no':
            return int(value)
        return value

    def context(self):
        if not 'order_no' in self.data:
            raise Exception("Need order no")
        self.super().context()

Then add the validator:

order_no = "12345"
DQ("(o.order_no, o.customer) Orders{o.order_no == '%(order_no)')} b")
    .validator(MyContextValidator)
    .context(locals())
    .tuples()

You can set your own validator class in Django settings:

DJAQ_VALIDATOR = MyValidator

The request parameter of the API view is added to the context and will be available to the validator as request.

Column expressions

Doing column arithmetic is supported directly in the query syntax:

DQ("""(b.name,
    b.price as price,
    0.2 as discount,
    b.price*0.2 as discount_price,
    b.price - (b.price*0.2) as diff
    ) Book b""")

You can use constants:

In [60]: list(DQ("(b.name, 'great read') Book b").limit(1).tuples())
Out[60]: [('Range total author impact.', 'great read')]

You can use the common operators and functions of your underlying db.

The usual arithmetic:

In [36]: list(DQ("(b.name, 1+1) Book b").limit(1).tuples())
Out[36]: [('Range total author impact.', 2)]
In [38]: list(DQ("(b.name, 2.0/4) Book b").limit(1).tuples())
Out[38]: [('Range total author impact.', Decimal('0.50000000000000000000'))]
In [44]: list(DQ("(2*3) Book b").limit(1).tuples())
Out[44]: [(6,)]

Modulo:

In [55]: list(DQ("(mod(4.0,3)) Book b").limit(1).tuples())
Out[55]: [(Decimal('1.0'),)]

Comparison as a boolean expression:

In [45]: list(DQ("(2 > 3) Book b").limit(1).tuples())
Out[45]: [(False,)]

While the syntax has a superficial resemblance to Python, you do not have access to any functions of the Python Standard Libary.

Subqueries and in clause

You can reference subqueries within a Djaq expression using

• Another DjangoQuery

• A Queryset

• A list

The two most useful cases are using a subquery in the filter condition:

DQ('(b.id, b.name) Book{b.id in ["(Book.id)"]} b')

And using a subquery in the selected columns expression:

DQ('(p.name, ["(count(b.id)) Book{Publisher.id == b.publisher} b"]) Publisher p')

You can use an IN clause with the keyword in (note lower case) If you are writing queries via the Python API. Create one DjangoQuery and reference it with @queryname:

DQ("(b.id) Book{name == 'B*'} b", name='dq_sub')
dq = DQ("(b.name, b.price) Book{id in '@dq_sub'} b")

Note that you have to pass a name to the DjangoQuery to reference it later. We can also use the data parameter to pass a QuerySet to the DjangoQuery:

qs = Book.objects.filter(name__startswith="B").only('id')
dq = DQ("(b.name, b.price) Book{id in '@qs_sub'} b", names={"qs_sub": qs})

qs = Book.objects.filter(name__startswith="B").only('id')
ids = [rec.id for rec in qs]
dq = DQ("(b.name, b.price) Book{id in '@qs_sub'} b", names={"qs_sub": ids})

As with QuerySets it is nearly always faster to generate a sub query than use an itemised list.

Order by

You can order_by like this:

DQ("(b.id) Book{b.price > 20} b order by (b.name)")

Descending order:

DQ("(b.id) Book{b.price > 20} b order by (-b.name)")

You can have multple order by expressions.

DQ("(b.name, Publisher.name) Book{b.price > 20} b order by (-b.name, b.publisher.name)")

Count

There are a couple ways to count results. These both return the exact same thing:

DQ("(Book.id)").count()

DQ("(count(Book.id)) Book").value()

Datetimes

Datetimes are provided as strings in the iso format that your backend expects, like ‘2019-01-01 18:00:00’.

Constants

None, True, False are replaced in SQL with NULL, TRUE, FALSE. All of the following work:

DQ("(b.id, b.name) Book{in_print is True} b")
DQ("(b.id, b.name) Book{in_print is not True} b")
DQ("(b.id, b.name) Book{in_print is False} b")
DQ("(b.id, b.name) Book{in_print == True} b")

Slicing

You cannot slice a DjangoQuery because this would frustrate a design goal of Djaq to provide the performance advantages of cursor-like behaviour.

You can use limit() and offset():

DjangoQuery("...").offset(1000).limit(100).tuples()

Which will provide you with the first hundred results starting from the 1000th record.

Rewind cursor

You can rewind the cursor but this is just executing the SQL again:

list(dq.tuples())

# now, calling `dq.tuples()` returns nothing

list(dq.rewind().tuples())

# you will again see results

If you call DjangoQuery.context(data), that will effectively rewind the cursor since an entirely new query is created and the implementation currently doesn’t care if data is the same context as previously supplied.

Schema

There is a function to get the schema available to a calling client:

from djaq.app_utils import get_schema
print(get_schema())

Pass the same whitelist you use for exposing the query endpoint:

wl = {"books": []}
print(get_schema(whitelist=wl))

Comparing to Django QuerySets

Djaq is not a replacement for Querysets. They have different purposes. The QuerySet API is not a remote API. You can use Djaq queries inside your Django application and Djaq syntax can be more intuitive and simpler than QuerySets. But Querysets are highly integrated with Django and have been developed over 15 years by many developers. Plus you get code completion in your IDE with QuerySets. It is a very well thought out framework that is the best choice working within a service based on Django’s ORM. You could probably write a complete transactional Django application with Djaq and not use QuerySets at all but you’d be going against the framework.

This section is intended to highlight differences for users with high familiarity with the QuerySet class for the purpose of understanding capabilities and limitations of DjangoQuery.

Django provides significant options for adjusting query generation to fit a specific use case, only(), select_related(), prefetch_related() are all useful for different cases. Here’s a point-by-point comparison with Djaq:

• only(): Djaq always works in “only” mode. Only explicitly requested fields are returned.

• select_related(): The select clause only returns those columns explicitly defined. This feature makes loading of related fields non-lazy. In contrast, queries are always non-lazy in Djaq.

• prefetch_related(): When you have a m2m field as a column expression, the model hosting that field is repeated in results as many times as necessary. Another way is to use a separate query for the m2m related records. In anycase, prefetch_related() this is not relevant in Djaq.

• F expressions: These are QuerySet workarounds for not being able to write expressions in the query for things like column value arithmetic and other expressions you want to have the db calculate. Djaq lets you write these directly and naturally as part of its syntax.

• To aggregate with Querysets, you use aggregate(), whereas Djaq aggregates results implicitly whenever an aggregate function appears in the column expressions.

• Model instances with QuerySets exactly represent the corresponding Django model. Djaq’s usual return formats, like dicts(), tuples(), etc. are more akin to QuerySet.value_list().

Let’s look at some direct query comparisons:

Get the average price of books:

DQ("(avg(b.price)) Book b")

compared to QuerySet:

Book.objects.all().aggregate(Avg('price'))

Get the difference from the average off the maximum price:

DQ("(Publisher.name, max(Book.price) - avg(Book.price) as price_diff) Book b")

compared to QuerySet:

Book.objects.aggregate(price_diff=Max('price', output_field=FloatField()) - Avg('price'))

Count books per publisher:

DQ("(Publisher.name, count(Book.id) as num_books) Book b")

compared to QuerySet:

Publisher.objects.annotate(num_books=Count("book"))

Count books with ratings up to and over 5:

DQ("""(sum(iif(b.rating < 5, b.rating, 0)) as below_5,
    sum(iif(b.rating >= 5, b.rating, 0)) as above_5)
    Book b""")

compared to QuerySet:

above_5 = Count('book', filter=Q(book__rating__gt=5))
below_5 = Count('book', filter=Q(book__rating__lte=5))
Publisher.objects.annotate(below_5=below_5).annotate(above_5=above_5)

Get average, maximum, minimum price of books:

DQ("(avg(b.price), max(b.price), min(b.price)) Book b")

compared to QuerySet:

Book.objects.aggregate(Avg('price'), Max('price'), Min('price'))

Just as there is a ModelInstance class in Django, we have a DQResult class:

objs(): return a DQResult for each result row, basically a namespace for the object:

dq = DQ("(b.id, b.name, Publisher.name as publisher) Book b")
for book in dq.objs():
    title = book.name
    publisher = book.publisher
    ...

Note that by default, you iterate using a generator. You cannot slice a generator.

Simple counts:

DjangoQuery.value(): when you know the result is a single row with a single value, you can immediately access it without further iterations:

DQ("(count(b.id)) Book b").value()

will return a single integer value representing the count of books.

Django Subquery and OuterRef

The following do pretty much the same thing:

# QuerySet
pubs = Publisher.objects.filter(pk=OuterRef('publisher')).only('pk')
Book.objects.filter(publisher__in=Subquery(pubs))

# Djaq
DQ("(p.id) Publisher p", name='pubs')
DQ("(b.name) Book{publisher in '@pubs'} b")

Obviously, in both cases, you would be filtering Publisher to make it actually useful, but the effect and verbosity can be extrapolated from the above.

Most importantly, sending a query request over the wire, you can reference the outer scope:

DQ('(p.name, ["(count(b.id)) Book{Publisher.id == b.publisher} b"]) Publisher p')

the subquery output expression references the outer scope. It evaluates to the following SQL:

SELECT
   "books_publisher"."name",
   (SELECT count("books_book"."id") FROM books_book WHERE "books_publisher"."id" = "books_book"."publisher_id")
FROM books_publisher

There are some constraints on using subqueries like this. For instance, the subquery cannot contain any joins.

Sample Project

If you want to use Djaq right away in your own test project and you feel confident, crack on. In that case skip the following instructions for using the sample Bookshop project. Or, if you want to try the sample project, clone the django repo:

git clone [e-mail]:paul-wolf/djaq.git
cd djaq/bookshop

If you clone the repo and use the sample project, you don’t need to include Djaq as a requirement because it’s included as a module by a softlink. Create the virtualenv:

python -m venv .venv

Activate the virtual environment:

source .venv/bin/activate

The module itself does not install Django and there are no further requirements. To install dependencies for the sample application:

pip install -r requirements.txt

Now make sure there is a Postgresql instance running. The settings are like this:

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.postgresql_psycopg2',
        'NAME': 'bookshop',
    },

So, it assumes peer authentication. Change to suit your needs. Now you can migrate. Make sure the virtualenv is activated!

./manage.py migrate

We provide a script to create some sample data:

./manage.py build_data --book-count 2000

This creates 2000 books and associated data.

The example app comes with a management command to run queries:

./manage.py djaq "(Publisher.name, max(Book.price) - round(avg(Book.price)) as diff) Book b"  --format json

Output of the command should look like this:

▶ ./manage.py djaq "(Publisher.name, max(Book.price) - round(avg(Book.price)) as diff) Book b"  --format json
SELECT books_publisher.name, (max(books_book.price) - round(avg(books_book.price))) FROM books_book LEFT JOIN books_publisher ON (books_book.publisher_id = books_publisher.id)  GROUP BY books_publisher.name LIMIT 10
{"publisher_name": "Avila, Garza and Ward", "diff": 14.0}
{"publisher_name": "Boyer-Clements", "diff": 16.0}
{"publisher_name": "Clark, Garza and York", "diff": 15.0}
{"publisher_name": "Clarke PLC", "diff": 14.0}
{"publisher_name": "Griffin-Blake", "diff": 16.0}
{"publisher_name": "Hampton-Davis", "diff": 13.0}
{"publisher_name": "Jones LLC", "diff": 15.0}
{"publisher_name": "Lane-Kim", "diff": 15.0}
{"publisher_name": "Norris-Bennett", "diff": 14.0}
{"publisher_name": "Singleton-King", "diff": 17.0}

Notice the SQL used to retrieve data is printed first.

The best approach now would be to trial various queries using the Djaq UI as explained above.

Finally, checkout the settings for the bookshop. You will notice that many admin models are not accessible. In a real application we’d want to prevent access to user data and other data on perhaps a finer grained level.