This package offers an easy to use GraphQL client. It's composed of the following modules:

• sgqlc.types: declare GraphQL in Python, base to generate and interpret queries. Submodule sgqlc.types.datetime will provide bindings for datetime and ISO 8601, while sgqlc.types.relay will expose Node, PageInfo and Connection.
• sgqlc.operation: use declared types to generate and interpret queries.
• sgqlc.endpoint: provide access to GraphQL endpoints, notably sgqlc.endpoint.http provides HTTPEndpoint using urllib.request.urlopen().

Straight from http://graphql.org:

A query language for your API

GraphQL is a query language for APIs and a runtime for fulfilling those queries with your existing data. GraphQL provides a complete and understandable description of the data in your API, gives clients the power to ask for exactly what they need and nothing more, makes it easier to evolve APIs over time, and enables powerful developer tools.

It was created by Facebook based on their problems and solutions using REST to develop applications to consume their APIs. It was publicly announced at React.js Conf 2015 and started to gain traction since then. Right now there are big names transitioning from REST to GraphQL: Yelp Shopify and GitHub, that did an excellent post to explain why they changed.

A short list of advantages over REST:

• Built-in schema, with documentation, strong typing and introspection. There is no need to use Swagger or any other external tools to play with it. Actually GraphQL provides a standard in-browser IDE for exploring GraphQL endpoints: https://github.com/graphql/graphiql;
• Only the fields that you want. The queries must explicitly select which fields are required, and that's all you're getting. If more fields are added to the type, they won't break the API, since the new fields won't be returned to old clients, as they didn't ask for such fields. This makes much easier to keep APIs stable and avoids versioning. Standard REST usually delivers all available fields in the results, and when new fields are to be included, a new API version is added (reflected in the URL path, or in an HTTP header);
• All data in one request. Instead of navigating hypermedia-driven RESTful services, like discovering new "_links": {"href"... and executing a new HTTP request, with GraphQL you specify nested queries and let the whole navigation be done by the server. This reduces latency a lot;
• The resulting JSON object matches the given query exactly; if you requested { parent { child { info } } }, you're going to receive the JSON object {"parent": {"child": {"info": value }}}.

From GitHub's Migrating from REST to GraphQL one can see these in real life:

$ curl -v https://api.github.com/orgs/github/members
[
  {
    "login": "...",
    "id": 1234,
    "avatar_url": "https://avatars3.githubusercontent.com/u/...",
    "gravatar_id": "",
    "url": "https://api.github.com/users/...",
    "html_url": "https://github.com/...",
    "followers_url": "https://api.github.com/users/.../followers",
    "following_url": "https://api.github.com/users/.../following{/other_user}",
    "gists_url": "https://api.github.com/users/.../gists{/gist_id}",
    "starred_url": "https://api.github.com/users/.../starred{/owner}{/repo}",
    "subscriptions_url": "https://api.github.com/users/.../subscriptions",
    "organizations_url": "https://api.github.com/users/.../orgs",
    "repos_url": "https://api.github.com/users/.../repos",
    "events_url": "https://api.github.com/users/.../events{/privacy}",
    "received_events_url": "https://api.github.com/users/.../received_events",
    "type": "User",
    "site_admin": true
  },
  ...
]

brings the whole set of member information, however you just want name and avatar URL:

query {
  organization(login:"github") { # select the organization
    members(first: 100) {        # then select the organization's members
      edges {  # edges + node: convention for paginated queries
        node {
          name
          avatarUrl
        }
      }
    }
  }
}

Likewise, instead of 4 HTTP requests:

curl -v https://api.github.com/repos/profusion/sgqlc/pulls/9
curl -v https://api.github.com/repos/profusion/sgqlc/pulls/9/commits
curl -v https://api.github.com/repos/profusion/sgqlc/issues/9/comments
curl -v https://api.github.com/repos/profusion/sgqlc/pulls/9/reviews

A single GraphQL query brings all the needed information, and just the needed information:

query {
  repository(owner: "profusion", name: "sgqlc") {
    pullRequest(number: 9) {
      commits(first: 10) { # commits of profusion/sgqlc PR #9
        edges {
          node { commit { oid, message } }
        }
      }
      comments(first: 10) { # comments of profusion/sgqlc PR #9
        edges {
          node {
            body
            author { login }
          }
        }
      }
      reviews(first: 10) { # reviews of profusion/sgqlc/ PR #9
        edges { node { state } }
      }
    }
  }
}

As seen above, writing GraphQL queries is very easy, and it is equally easy to interpret the results. So what was the rationale to create sgqlc?

• GraphQL has its domain-specific language (DSL), and mixing two languages is always painful, as seen with SQL + Python, HTML + Python... Being able to write just Python in Python is much better. Not to say that GraphQL naming convention is closer to Java/JavaScript, using aNameFormat instead of Python's a_name_format.
• Navigating dict-of-stuff is a bit painful: d["repository"]["pullRequest"]["commits"]["edges"]["node"], since these are valid Python identifiers, we better write: repository.pull_request.commits.edges.node.
• Handling new scalar types. GraphQL allows one to define new scalar types, such as Date, Time and DateTime. Often these are serialized as ISO 8601 strings and the user must parse them in their application. We offer sgqlc.types.datetime to automatically generate datetime.date, datetime.time and datetime.datetime.
• Make it easy to write dynamic queries, including nested. As seen, GraphQL can be used to fetch lots of information in one go; however if what you need (arguments and fields) changes based on some variable, such as user input or cached data, then you need to concatenate strings to compose the final query. This can be error prone and servers may block you due to invalid queries. Some tools "solve" this by parsing the query locally before sending it to server. However usually the indentation is screwed and reviewing it is painful. We change that approach: use sgqlc.operation.Operation and it will always generate valid queries, which can be printed out and properly indented. Bonus point is that it can be used to later interpret the JSON results into native Python objects.
• Usability improvements whenever needed. For instance Relay published their Cursor Connections Specification and its widely used. To load more data, you need to extend the previous data with newly fetched information, updating not only the nodes and edges, but also page information. This is done automatically by sgqlc.types.relay.Connection.

It also helps with code-generation, sgqlc-codegen can generate both the classes matching a GraphQL Schema or functions to return sgqlc.operation.Operation based on executable documents GraphQL Domain Specific Language (DSL).

Automatic:

pip install sgqlc

From source using pip:

pip install .

To reach a GraphQL endpoint using synchronous HTTPEndpoint with a hand-written query (see more at examples/basic/01_http_endpoint.py):

from sgqlc.endpoint.http import HTTPEndpoint

url = 'http://server.com/graphql'
headers = {'Authorization': 'bearer TOKEN'}

query = 'query { ... }'
variables = {'varName': 'value'}

endpoint = HTTPEndpoint(url, headers)
data = endpoint(query, variables)

To make an asynchronous query, use the httpx library with HTTPXEndpoint (see more at examples/basic/04_httpx_endpoint.py):

from sgqlc.endpoint.httpx import HTTPXEndpoint
import httpx

async def main():
    url = 'http://server.com/graphql'
    headers = {'Authorization': 'bearer TOKEN'}

    query = 'query { ... }'
    variables = {'varName': 'value'}

    endpoint = HTTPXEndpoint(url, headers, client=httpx.AsyncClient())
    data = await endpoint(query, variables)

However, writing GraphQL queries and later interpreting the results may be cumbersome. That's solved by our sgqlc.types, which is usually paired with sgqlc.operation to generate queries and then interpret results (see more at examples/basic/02_schema_types.py). The example below matches a subset of GitHub API v4. In GraphQL syntax it would be:

query {
  repository(owner: "profusion", name: "sgqlc") {
    issues(first: 100) {
      nodes {
        number
        title
      }
      pageInfo {
        hasNextPage
        endCursor
      }
    }
  }
}

The output JSON object is:

{
  "data": {
    "repository": {
      "issues": {
        "nodes": [
          {"number": 1, "title": "..."},
          {"number": 2, "title": "..."}
        ]
      },
      "pageInfo": {
         "hasNextPage": false,
         "endCursor": "..."
      }
    }
  }
}

from sgqlc.endpoint.http import HTTPEndpoint
from sgqlc.types import Type, Field, list_of
from sgqlc.types.relay import Connection, connection_args
from sgqlc.operation import Operation

# Declare types matching GitHub GraphQL schema:
class Issue(Type):
    number = int
    title = str

class IssueConnection(Connection):  # Connection provides page_info!
    nodes = list_of(Issue)

class Repository(Type):
    issues = Field(IssueConnection, args=connection_args())

class Query(Type):  # GraphQL's root
    repository = Field(Repository, args={'owner': str, 'name': str})

# Generate an operation on Query, selecting fields:
op = Operation(Query)
# select a field, here with selection arguments, then another field:
issues = op.repository(owner=owner, name=name).issues(first=100)
# select sub-fields explicitly: { nodes { number title } }
issues.nodes.number()
issues.nodes.title()
# here uses __fields__() to select by name (*args)
issues.page_info.__fields__('has_next_page')
# here uses __fields__() to select by name (**kwargs)
issues.page_info.__fields__(end_cursor=True)

# you can print the resulting GraphQL
print(op)

# Call the endpoint:
data = endpoint(op)

# Interpret results into native objects
repo = (op + data).repository
for issue in repo.issues.nodes:
    print(issue)

Since we don't want to clobber GraphQL fields, we cannot provide nicely named methods. Therefore we use overloaded methods such as __iadd__, __add__, __bytes__ (compressed GraphQL representation) and __str__ (indented GraphQL representation).

To select fields by name, use __fields__(*names, **names_and_args). This helps with repetitive situations and can be used to "include all fields", or "include all except...":

# just 'a' and 'b'
type_selection.__fields__('a', 'b')
type_selection.__fields__(a=True, b=True) # equivalent

# a(arg1: value1), b(arg2: value2):
type_selection.__fields__(
    a={'arg1': value1},
    b={'arg2': value2})

# selects all possible fields
type_selection.__fields__()

# all but 'a' and 'b'
type_selection.__fields__(__exclude__=('a', 'b'))
type_selection.__fields__(a=False, b=False)

Manually converting an existing GraphQL schema to sgqlc.types subclasses is boring and error prone. To aid such task we offer a code generator that outputs a Python module straight from JSON of an introspection call:

user@host$ python3 -m sgqlc.introspection \
     --exclude-deprecated \
     --exclude-description \
     -H "Authorization: bearer ${GH_TOKEN}" \
     https://api.github.com/graphql \
     github_schema.json
user@host$ sgqlc-codegen schema github_schema.json github_schema.py

This generates github_schema that provides the sgqlc.types.Schema instance of the same name github_schema. Then it's a matter of using that in your Python code, as in the example below from examples/github/github_agile_dashboard.py:

from sgqlc.operation import Operation
from github_schema import github_schema as schema

op = Operation(schema.Query)  # note 'schema.'

# -- code below follows as the original usage example:

# select a field, here with selection arguments, then another field:
issues = op.repository(owner=owner, name=name).issues(first=100)
# select sub-fields explicitly: { nodes { number title } }
issues.nodes.number()
issues.nodes.title()
# here uses __fields__() to select by name (*args)
issues.page_info.__fields__('has_next_page')
# here uses __fields__() to select by name (**kwargs)
issues.page_info.__fields__(end_cursor=True)

# you can print the resulting GraphQL
print(op)

# Call the endpoint:
data = endpoint(op)

# Interpret results into native objects
repo = (op + data).repository
for issue in repo.issues.nodes:
    print(issue)

You can also generate these operations given a GraphQL Domain Specific Language (DSL) operation:

# sample_operations.gql

query ListIssues($owner: String!, $name: String!) {
    repository(owner: $owner, name: $name) {
        issues(first: 100) {
            nodes {
                number
                title
            }
            pageInfo {
                hasNextPage
                endCursor
            }
        }
    }
}

user@host$ sgqlc-codegen operation \
   --schema github_schema.json \
   github_schema \
   sample_operations.py \
   sample_operations.gql

This generates sample_operations.py that provides the Operation. Then it's a matter of using that in your Python code, as in the example below from examples/github/github-agile-dashboard.py:

from sample_operations import Operations

op = Operations.query.list_issues

# you can print the resulting GraphQL
print(op)

# Call the endpoint:
data = endpoint(op, {'owner': owner, 'name': name})

# Interpret results into native objects
repo = (op + data).repository
for issue in repo.issues.nodes:
    print(issue)

• Gustavo Sverzut Barbieri

sgqlc is licensed under the ISC.

You need to use poetry.

poetry install --all-extras --with dev
poetry shell

Install the pre-commit:

pre-commit install -f

Run the tests (one of the below):

pre-commit run -a            # run all tests: flake8, pytest, ...
pre-commit run -a flake8     # run only flake8
pre-commit run -a tests      # run only pytest (unit tests)

Keep 100% coverage. You can look at the coverage report at cover/index.html. To do that, prefer doctest so it serves as both documentation and test. However we use pytest to write explicit tests that would be hard to express using doctest.

Build and review the generated Sphinx documentation, and validate if your changes look right:

sphinx-build doc/source doc/build
open doc/build/html/index.html

To integrate changes from another branch, please rebase instead of creating merge commits ( read more).

The following repositories provides public schemas generated using sgqlc-codegen:

• Mogost/sgqlc-schemas GitHub, Monday.com