Filters

• Filters For Formatting Data
• Forcing Variables To Be Defined
• Defaulting Undefined Variables
• Omitting Parameters
• List Filters
• Set Theory Filters
• Random Number Filter
• Shuffle Filter
• Math
• JSON Query Filter
• IP address filter
• Network CLI filters
• Network XML filters
• Hashing filters
• Combining hashes/dictionaries
• Extracting values from containers
• Comment Filter
• URL Split Filter
• Regular Expression Filters
• Other Useful Filters
• Combination Filters
• Debugging Filters

Filters in Ansible are from Jinja2, and are used for transforming data inside a template expression. Jinja2 ships with many filters. See builtin filters in the official Jinja2 template documentation.

Take into account that templating happens on the Ansible controller, not on the task’s target host, so filters also execute on the controller as they manipulate local data.

In addition the ones provided by Jinja2, Ansible ships with it’s own and allows users to add their own custom filters.

Filters For Formatting Data

The following filters will take a data structure in a template and render it in a slightly different format. These are occasionally useful for debugging:

{{ some_variable | to_json }}
{{ some_variable | to_yaml }}

For human readable output, you can use:

{{ some_variable | to_nice_json }}
{{ some_variable | to_nice_yaml }}

It’s also possible to change the indentation of both (new in version 2.2):

{{ some_variable | to_nice_json(indent=2) }}
{{ some_variable | to_nice_yaml(indent=8) }}

Alternatively, you may be reading in some already formatted data:

{{ some_variable | from_json }}
{{ some_variable | from_yaml }}

for example:

tasks:
  - shell: cat /some/path/to/file.json
    register: result

  - set_fact:
      myvar: "{{ result.stdout | from_json }}"

Forcing Variables To Be Defined

The default behavior from ansible and ansible.cfg is to fail if variables are undefined, but you can turn this off.

This allows an explicit check with this feature off:

{{ variable | mandatory }}

The variable value will be used as is, but the template evaluation will raise an error if it is undefined.

Defaulting Undefined Variables

Jinja2 provides a useful ‘default’ filter that is often a better approach to failing if a variable is not defined:

{{ some_variable | default(5) }}

In the above example, if the variable ‘some_variable’ is not defined, the value used will be 5, rather than an error being raised.

If the variable evaluates to an empty string, the second parameter of the filter should be set to true:

{{ lookup('env', 'MY_USER') | default('admin', true) }}

Omitting Parameters

As of Ansible 1.8, it is possible to use the default filter to omit module parameters using the special omit variable:

- name: touch files with an optional mode
  file: dest={{ item.path }} state=touch mode={{ item.mode | default(omit) }}
  loop:
    - path: /tmp/foo
    - path: /tmp/bar
    - path: /tmp/baz
      mode: "0444"

For the first two files in the list, the default mode will be determined by the umask of the system as the mode= parameter will not be sent to the file module while the final file will receive the mode=0444 option.

List Filters

These filters all operate on list variables.

To get the minimum value from list of numbers:

{{ list1 | min }}

To get the maximum value from a list of numbers:

{{ [3, 4, 2] | max }}

Flatten a list (same thing the flatten lookup does):

{{ [3, [4, 2] ] | flatten }}

Flatten only the first level of a list (akin to the items lookup):

{{ [3, [4, [2]] ] | flatten(level=1) }}

To get the minimum value from list of numbers:

.. _set_theory_filters:

Set Theory Filters

All these functions return a unique set from sets or lists.

To get a unique set from a list:

{{ list1 | unique }}

To get a union of two lists:

{{ list1 | union(list2) }}

To get the intersection of 2 lists (unique list of all items in both):

{{ list1 | intersect(list2) }}

To get the difference of 2 lists (items in 1 that don’t exist in 2):

{{ list1 | difference(list2) }}

To get the symmetric difference of 2 lists (items exclusive to each list):

{{ list1 | symmetric_difference(list2) }}

Random Number Filter

This filter can be used similar to the default jinja2 random filter (returning a random item from a sequence of items), but can also generate a random number based on a range.

To get a random item from a list:

"{{ ['a','b','c'] | random }}"
# => 'c'

To get a random number between 0 and a specified number:

"{{ 60 | random }} * * * * root /script/from/cron"
# => '21 * * * * root /script/from/cron'

Get a random number from 0 to 100 but in steps of 10:

{{ 101 | random(step=10) }}
# => 70

Get a random number from 1 to 100 but in steps of 10:

{{ 101 | random(1, 10) }}
# => 31
{{ 101 | random(start=1, step=10) }}
# => 51

As of Ansible version 2.3, it’s also possible to initialize the random number generator from a seed. This way, you can create random-but-idempotent numbers:

"{{ 60 | random(seed=inventory_hostname) }} * * * * root /script/from/cron"

Shuffle Filter

This filter will randomize an existing list, giving a different order every invocation.

To get a random list from an existing list:

{{ ['a','b','c'] | shuffle }}
# => ['c','a','b']
{{ ['a','b','c'] | shuffle }}
# => ['b','c','a']

As of Ansible version 2.3, it’s also possible to shuffle a list idempotent. All you need is a seed.:

{{ ['a','b','c'] | shuffle(seed=inventory_hostname) }}
# => ['b','a','c']

note that when used with a non ‘listable’ item it is a noop, otherwise it always returns a list

Math

Get the logarithm (default is e):

{{ myvar | log }}

Get the base 10 logarithm:

{{ myvar | log(10) }}

Give me the power of 2! (or 5):

{{ myvar | pow(2) }}
{{ myvar | pow(5) }}

Square root, or the 5th:

{{ myvar | root }}
{{ myvar | root(5) }}

Note that jinja2 already provides some like abs() and round().

JSON Query Filter

Sometimes you end up with a complex data structure in JSON format and you need to extract only a small set of data within it. The json_query filter lets you query a complex JSON structure and iterate over it using a loop structure.

Now, let’s take the following data structure:

domain_definition:
    domain:
        cluster:
            - name: "cluster1"
            - name: "cluster2"
        server:
            - name: "server11"
              cluster: "cluster1"
              port: "8080"
            - name: "server12"
              cluster: "cluster1"
              port: "8090"
            - name: "server21"
              cluster: "cluster2"
              port: "9080"
            - name: "server22"
              cluster: "cluster2"
              port: "9090"
        library:
            - name: "lib1"
              target: "cluster1"
            - name: "lib2"
              target: "cluster2"

To extract all clusters from this structure, you can use the following query:

- name: "Display all cluster names"
  debug:
    var: item
  loop: "{{ domain_definition | json_query('domain.cluster[*].name') }}"

Same thing for all server names:

- name: "Display all server names"
  debug:
    var: item
  loop: "{{ domain_definition | json_query('domain.server[*].name') }}"

This example shows ports from cluster1:

- name: "Display all ports from cluster1"
  debug:
    var: item
  loop: "{{ domain_definition | json_query(server_name_cluster1_query) }}"
  vars:
    server_name_cluster1_query: "domain.server[?cluster=='cluster1'].port"

Or, alternatively print out the ports in a comma separated string:

- name: "Display all ports from cluster1 as a string"
  debug:
    msg: "{{ domain_definition | json_query('domain.server[?cluster==`cluster1`].port') | join(', ') }}"

Or, using YAML single quote escaping:

- name: "Display all ports from cluster1"
  debug:
    var: item
  loop: "{{ domain_definition | json_query('domain.server[?cluster==''cluster1''].port') }}"

In this example, we get a hash map with all ports and names of a cluster:

- name: "Display all server ports and names from cluster1"
  debug:
    var: item
  loop: "{{ domain_definition | json_query(server_name_cluster1_query) }}"
  vars:
    server_name_cluster1_query: "domain.server[?cluster=='cluster2'].{name: name, port: port}"

IP address filter

To test if a string is a valid IP address:

{{ myvar | ipaddr }}

You can also require a specific IP protocol version:

{{ myvar | ipv4 }}
{{ myvar | ipv6 }}

IP address filter can also be used to extract specific information from an IP address. For example, to get the IP address itself from a CIDR, you can use:

{{ '192.0.2.1/24' | ipaddr('address') }}

More information about ipaddr filter and complete usage guide can be found in ipaddr filter.

Network CLI filters

To convert the output of a network device CLI command into structured JSON output, use the parse_cli filter:

{{ output | parse_cli('path/to/spec') }}

The parse_cli filter will load the spec file and pass the command output through it, returning JSON output. The YAML spec file defines how to parse the CLI output.

The spec file should be valid formatted YAML. It defines how to parse the CLI output and return JSON data. Below is an example of a valid spec file that will parse the output from the show vlan command.

---
vars:
  vlan:
    vlan_id: "{{ item.vlan_id }}"
    name: "{{ item.name }}"
    enabled: "{{ item.state != 'act/lshut' }}"
    state: "{{ item.state }}"

keys:
  vlans:
    value: "{{ vlan }}"
    items: "^(?P<vlan_id>\\d+)\\s+(?P<name>\\w+)\\s+(?P<state>active|act/lshut|suspended)"
  state_static:
    value: present

The spec file above will return a JSON data structure that is a list of hashes with the parsed VLAN information.

The same command could be parsed into a hash by using the key and values directives. Here is an example of how to parse the output into a hash value using the same show vlan command.

---
vars:
  vlan:
    key: "{{ item.vlan_id }}"
    values:
      vlan_id: "{{ item.vlan_id }}"
      name: "{{ item.name }}"
      enabled: "{{ item.state != 'act/lshut' }}"
      state: "{{ item.state }}"

keys:
  vlans:
    value: "{{ vlan }}"
    items: "^(?P<vlan_id>\\d+)\\s+(?P<name>\\w+)\\s+(?P<state>active|act/lshut|suspended)"
  state_static:
    value: present

Another common use case for parsing CLI commands is to break a large command into blocks that can be parsed. This can be done using the start_block and end_block directives to break the command into blocks that can be parsed.

---
vars:
  interface:
    name: "{{ item[0].match[0] }}"
    state: "{{ item[1].state }}"
    mode: "{{ item[2].match[0] }}"

keys:
  interfaces:
    value: "{{ interface }}"
    start_block: "^Ethernet.*$"
    end_block: "^$"
    items:
      - "^(?P<name>Ethernet\\d\\/\\d*)"
      - "admin state is (?P<state>.+),"
      - "Port mode is (.+)"

The example above will parse the output of show interface into a list of hashes.

The network filters also support parsing the output of a CLI command using the TextFSM library. To parse the CLI output with TextFSM use the following filter:

{{ output.stdout[0] | parse_cli_textfsm('path/to/fsm') }}

Use of the TextFSM filter requires the TextFSM library to be installed.

Network XML filters

To convert the XML output of a network device command into structured JSON output, use the parse_xml filter:

{{ output | parse_xml('path/to/spec') }}

The parse_xml filter will load the spec file and pass the command output through formatted as JSON.

The spec file should be valid formatted YAML. It defines how to parse the XML output and return JSON data.

Below is an example of a valid spec file that will parse the output from the show vlan | display xml command.

---
vars:
  vlan:
    vlan_id: "{{ item.vlan_id }}"
    name: "{{ item.name }}"
    desc: "{{ item.desc }}"
    enabled: "{{ item.state.get('inactive') != 'inactive' }}"
    state: "{% if item.state.get('inactive') == 'inactive'%} inactive {% else %} active {% endif %}"

keys:
  vlans:
    value: "{{ vlan }}"
    top: configuration/vlans/vlan
    items:
      vlan_id: vlan-id
      name: name
      desc: description
      state: ".[@inactive='inactive']"

The spec file above will return a JSON data structure that is a list of hashes with the parsed VLAN information.

The same command could be parsed into a hash by using the key and values directives. Here is an example of how to parse the output into a hash value using the same show vlan | display xml command.

---
vars:
  vlan:
    key: "{{ item.vlan_id }}"
    values:
        vlan_id: "{{ item.vlan_id }}"
        name: "{{ item.name }}"
        desc: "{{ item.desc }}"
        enabled: "{{ item.state.get('inactive') != 'inactive' }}"
        state: "{% if item.state.get('inactive') == 'inactive'%} inactive {% else %} active {% endif %}"

keys:
  vlans:
    value: "{{ vlan }}"
    top: configuration/vlans/vlan
    items:
      vlan_id: vlan-id
      name: name
      desc: description
      state: ".[@inactive='inactive']"

The value of top is the XPath relative to the XML root node. In the example XML output given below, the value of top is configuration/vlans/vlan, which is an XPath expression relative to the root node (<rpc-reply>). configuration in the value of top is the outer most container node, and vlan is the inner-most container node.

items is a dictionary of key-value pairs that map user-defined names to XPath expressions that select elements. The Xpath expression is relative to the value of the XPath value contained in top. For example, the vlan_id in the spec file is a user defined name and its value vlan-id is the relative to the value of XPath in top

Attributes of XML tags can be extracted using XPath expressions. The value of state in the spec is an XPath expression used to get the attributes of the vlan tag in output XML.:

<rpc-reply>
  <configuration>
    <vlans>
      <vlan inactive="inactive">
       <name>vlan-1</name>
       <vlan-id>200</vlan-id>
       <description>This is vlan-1</description>
      </vlan>
    </vlans>
  </configuration>
</rpc-reply>

Hashing filters

To get the sha1 hash of a string:

{{ 'test1' | hash('sha1') }}

To get the md5 hash of a string:

{{ 'test1' | hash('md5') }}

Get a string checksum:

{{ 'test2' | checksum }}

Other hashes (platform dependent):

{{ 'test2' | hash('blowfish') }}

To get a sha512 password hash (random salt):

{{ 'passwordsaresecret' | password_hash('sha512') }}

To get a sha256 password hash with a specific salt:

{{ 'secretpassword' | password_hash('sha256', 'mysecretsalt') }}

An idempotent method to generate unique hashes per system is to use a salt that is consistent between runs:

{{ 'secretpassword' | password_hash('sha512', 65534 | random(seed=inventory_hostname) | string) }}

Hash types available depend on the master system running ansible, ‘hash’ depends on hashlib password_hash depends on passlib (http://passlib.readthedocs.io/en/stable/lib/passlib.hash.html).

Combining hashes/dictionaries

The combine filter allows hashes to be merged. For example, the following would override keys in one hash:

{{ {'a':1, 'b':2} | combine({'b':3}) }}

The resulting hash would be:

{'a':1, 'b':3}

The filter also accepts an optional recursive=True parameter to not only override keys in the first hash, but also recurse into nested hashes and merge their keys too

{{ {'a':{'foo':1, 'bar':2}, 'b':2} | combine({'a':{'bar':3, 'baz':4}}, recursive=True) }}

This would result in:

{'a':{'foo':1, 'bar':3, 'baz':4}, 'b':2}

The filter can also take multiple arguments to merge:

{{ a | combine(b, c, d) }}

In this case, keys in d would override those in c, which would override those in b, and so on.

This behaviour does not depend on the value of the hash_behaviour setting in ansible.cfg.

Extracting values from containers

The extract filter is used to map from a list of indices to a list of values from a container (hash or array):

{{ [0,2] | map('extract', ['x','y','z']) | list }}
{{ ['x','y'] | map('extract', {'x': 42, 'y': 31}) | list }}

The results of the above expressions would be:

['x', 'z']
[42, 31]

The filter can take another argument:

{{ groups['x'] | map('extract', hostvars, 'ec2_ip_address') | list }}

This takes the list of hosts in group ‘x’, looks them up in hostvars, and then looks up the ec2_ip_address of the result. The final result is a list of IP addresses for the hosts in group ‘x’.

The third argument to the filter can also be a list, for a recursive lookup inside the container:

{{ ['a'] | map('extract', b, ['x','y']) | list }}

This would return a list containing the value of b[‘a’][‘x’][‘y’].

Comment Filter

The comment filter allows to decorate the text with a chosen comment style. For example the following:

{{ "Plain style (default)" | comment }}

will produce this output:

#
# Plain style (default)
#

Similar way can be applied style for C (//...), C block (/*...*/), Erlang (%...) and XML (<!--...-->):

{{ "C style" | comment('c') }}
{{ "C block style" | comment('cblock') }}
{{ "Erlang style" | comment('erlang') }}
{{ "XML style" | comment('xml') }}

If you need a specific comment character that is not included by any of the above, you can customize it with:

{{ "My Special Case" | comment(decoration="! ") }}

producing:

!
! My Special Case
!

It is also possible to fully customize the comment style:

{{ "Custom style" | comment('plain', prefix='#######\n#', postfix='#\n#######\n   ###\n    #') }}

That will create the following output:

#######
#
# Custom style
#
#######
   ###
    #

The filter can also be applied to any Ansible variable. For example to make the output of the ansible_managed variable more readable, we can change the definition in the ansible.cfg file to this:

[defaults]

ansible_managed = This file is managed by Ansible.%n
  template: {file}
  date: %Y-%m-%d %H:%M:%S
  user: {uid}
  host: {host}

and then use the variable with the comment filter:

{{ ansible_managed | comment }}

which will produce this output:

#
# This file is managed by Ansible.
#
# template: /home/ansible/env/dev/ansible_managed/roles/role1/templates/test.j2
# date: 2015-09-10 11:02:58
# user: ansible
# host: myhost
#

URL Split Filter

The urlsplit filter extracts the fragment, hostname, netloc, password, path, port, query, scheme, and username from an URL. With no arguments, returns a dictionary of all the fields:

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('hostname') }}
# => 'www.acme.com'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('netloc') }}
# => 'user:[email protected]:9000'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('username') }}
# => 'user'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('password') }}
# => 'password'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('path') }}
# => '/dir/index.html'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('port') }}
# => '9000'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('scheme') }}
# => 'http'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('query') }}
# => 'query=term'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit('fragment') }}
# => 'fragment'

{{ "http://user:[email protected]:9000/dir/index.html?query=term#fragment" | urlsplit }}
# =>
#   {
#       "fragment": "fragment",
#       "hostname": "www.acme.com",
#       "netloc": "user:[email protected]:9000",
#       "password": "password",
#       "path": "/dir/index.html",
#       "port": 9000,
#       "query": "query=term",
#       "scheme": "http",
#       "username": "user"
#   }

Regular Expression Filters

To search a string with a regex, use the “regex_search” filter:

# search for "foo" in "foobar"
{{ 'foobar' | regex_search('(foo)') }}

# will return empty if it cannot find a match
{{ 'ansible' | regex_search('(foobar)') }}

To search for all occurrences of regex matches, use the “regex_findall” filter:

# Return a list of all IPv4 addresses in the string
{{ 'Some DNS servers are 8.8.8.8 and 8.8.4.4' | regex_findall('\b(?:[0-9]{1,3}\.){3}[0-9]{1,3}\b') }}

To replace text in a string with regex, use the “regex_replace” filter:

# convert "ansible" to "able"
{{ 'ansible' | regex_replace('^a.*i(.*)$', 'a\\1') }}

# convert "foobar" to "bar"
{{ 'foobar' | regex_replace('^f.*o(.*)$', '\\1') }}

# convert "localhost:80" to "localhost, 80" using named groups
{{ 'localhost:80' | regex_replace('^(?P<host>.+):(?P<port>\\d+)$', '\\g<host>, \\g<port>') }}

# convert "localhost:80" to "localhost"
{{ 'localhost:80' | regex_replace(':80') }}

To escape special characters within a regex, use the “regex_escape” filter:

# convert '^f.*o(.*)$' to '\^f\.\*o\(\.\*\)\$'
{{ '^f.*o(.*)$' | regex_escape() }}

Other Useful Filters

To add quotes for shell usage:

- shell: echo {{ string_value | quote }}

To use one value on true and another on false (new in version 1.9):

{{ (name == "John") | ternary('Mr','Ms') }}

To concatenate a list into a string:

{{ list | join(" ") }}

To get the last name of a file path, like ‘foo.txt’ out of ‘/etc/asdf/foo.txt’:

{{ path | basename }}

To get the last name of a windows style file path (new in version 2.0):

{{ path | win_basename }}

To separate the windows drive letter from the rest of a file path (new in version 2.0):

{{ path | win_splitdrive }}

To get only the windows drive letter:

{{ path | win_splitdrive | first }}

To get the rest of the path without the drive letter:

{{ path | win_splitdrive | last }}

To get the directory from a path:

{{ path | dirname }}

To get the directory from a windows path (new version 2.0):

{{ path | win_dirname }}

To expand a path containing a tilde (~) character (new in version 1.5):

{{ path | expanduser }}

To get the real path of a link (new in version 1.8):

{{ path | realpath }}

To get the relative path of a link, from a start point (new in version 1.7):

{{ path | relpath('/etc') }}

To get the root and extension of a path or filename (new in version 2.0):

# with path == 'nginx.conf' the return would be ('nginx', '.conf')
{{ path | splitext }}

To work with Base64 encoded strings:

{{ encoded | b64decode }}
{{ decoded | b64encode }}

To create a UUID from a string (new in version 1.9):

{{ hostname | to_uuid }}

To cast values as certain types, such as when you input a string as “True” from a vars_prompt and the system doesn’t know it is a boolean value:

- debug:
    msg: test
  when: some_string_value | bool

To make use of one attribute from each item in a list of complex variables, use the “map” filter (see the Jinja2 map() docs for more):

# get a comma-separated list of the mount points (e.g. "/,/mnt/stuff") on a host
{{ ansible_mounts | map(attribute='mount') | join(',') }}

To get date object from string use the to_datetime filter, (new in version in 2.2):

# Get total amount of seconds between two dates. Default date format is %Y-%m-%d %H:%M:%S but you can pass your own format
{{ (("2016-08-14 20:00:12" | to_datetime) - ("2015-12-25" | to_datetime('%Y-%m-%d'))).total_seconds()  }}

# Get remaining seconds after delta has been calculated. NOTE: This does NOT convert years, days, hours, etc to seconds. For that, use total_seconds()
{{ (("2016-08-14 20:00:12" | to_datetime) - ("2016-08-14 18:00:00" | to_datetime)).seconds  }}
# This expression evaluates to "12" and not "132". Delta is 2 hours, 12 seconds

# get amount of days between two dates. This returns only number of days and discards remaining hours, minutes, and seconds
{{ (("2016-08-14 20:00:12" | to_datetime) - ("2015-12-25" | to_datetime('%Y-%m-%d'))).days  }}

Combination Filters

This set of filters returns a list of combined lists. To get permutations of a list:

- name: give me largest permutations (order matters)
  debug:
    msg: "{{ [1,2,3,4,5] | permutations | list }}"

- name: give me permutations of sets of three
  debug:
    msg: "{{ [1,2,3,4,5] | permutations(3) | list }}"

Combinations always require a set size:

- name: give me combinations for sets of two
  debug:
    msg: "{{ [1,2,3,4,5] | combinations(2) | list }}"

To get a list combining the elements of other lists use zip:

- name: give me list combo of two lists
  debug:
   msg: "{{ [1,2,3,4,5] | zip(['a','b','c','d','e','f']) | list }}"

- name: give me shortest combo of two lists
  debug:
    msg: "{{ [1,2,3] | zip(['a','b','c','d','e','f']) | list }}"

To always exhaust all list use zip_longest:

- name: give me longest combo of three lists , fill with X
  debug:
    msg: "{{ [1,2,3] | zip_longest(['a','b','c','d','e','f'], [21, 22, 23], fillvalue='X') | list }}"

To format a date using a string (like with the shell date command), use the “strftime” filter:

# Display year-month-day
{{ '%Y-%m-%d' | strftime }}

# Display hour:min:sec
{{ '%H:%M:%S' | strftime }}

# Use ansible_date_time.epoch fact
{{ '%Y-%m-%d %H:%M:%S' | strftime(ansible_date_time.epoch) }}

# Use arbitrary epoch value
{{ '%Y-%m-%d' | strftime(0) }}          # => 1970-01-01
{{ '%Y-%m-%d' | strftime(1441357287) }} # => 2015-09-04

Debugging Filters

Use the type_debug filter to display the underlying Python type of a variable. This can be useful in debugging in situations where you may need to know the exact type of a variable:

{{ myvar | type_debug }}

A few useful filters are typically added with each new Ansible release. The development documentation shows how to extend Ansible filters by writing your own as plugins, though in general, we encourage new ones to be added to core so everyone can make use of them.