read_csv: new module to read CSV files (#49578)

* read_csv: new module to read CSV files

* Add a doc reference to the csvfile lookup plugin

* Enable the use of custom dialect options

* Improve error handling

* Fix PEP8

* Fix more PEP8

* Simplify custom dialect code

* Add integration tests

* Fixes for CI

* Fix for python 2.6
pull/49714/head
Dag Wieers 6 years ago committed by John R Barker
parent e25dac946f
commit 23ae3aa32a

@ -0,0 +1,235 @@
#!/usr/bin/python
# -*- coding: utf-8 -*-
# Copyright: (c) 2018, Dag Wieers (@dagwieers) <dag@wieers.com>
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt)
from __future__ import absolute_import, division, print_function
__metaclass__ = type
ANSIBLE_METADATA = {'metadata_version': '1.1',
'status': ['preview'],
'supported_by': 'community'}
DOCUMENTATION = r'''
---
module: read_csv
version_added: '2.8'
short_description: Read a CSV file
description:
- Read a CSV file and return a list or a dictionary, containing one dictionary per row.
author:
- Dag Wieers (@dagwieers)
options:
path:
description:
- The CSV filename to read data from.
type: str
required: yes
aliases: [ filename ]
key:
description:
- The column name used as a key for the resulting dictionary.
- If C(key) is unset, the module returns a list of dictionaries,
where each dictionary is a row in the CSV file.
type: str
dialect:
description:
- The CSV dialect to use when parsing the CSV file.
- Possible values include C(excel), C(excel-tab) or C(unix).
type: str
default: excel
fieldnames:
description:
- A list of field names for every column.
- This is needed if the CSV does not have a header.
type: list
unique:
description:
- Whether the C(key) used is expected to be unique.
type: bool
default: yes
delimiter:
description:
- A one-character string used to separate fields.
- When using this parameter, you change the default value used by C(dialect).
- The default value depends on the dialect used.
type: str
skipinitialspace:
description:
- Whether to ignore any whitespaces immediately following the delimiter.
- When using this parameter, you change the default value used by C(dialect).
- The default value depends on the dialect used.
type: bool
strict:
description:
- Whether to raise an exception on bad CSV input.
- When using this parameter, you change the default value used by C(dialect).
- The default value depends on the dialect used.
type: bool
notes:
- Ansible also ships with the C(csvfile) lookup plugin, which can be used to do selective lookups in CSV files from Jinja.
'''
EXAMPLES = r'''
# Example CSV file with header
#
# name,uid,gid
# dag,500,500
# jeroen,501,500
# Read a CSV file and access user 'dag'
- name: Read users from CSV file and return a dictionary
read_csv:
path: users.csv
key: name
register: users
delegate_to: localhost
- debug:
msg: 'User {{ users.dict.dag.name }} has UID {{ users.dict.dag.uid }} and GID {{ users.dict.dag.gid }}'
# Read a CSV file and access the first item
- name: Read users from CSV file and return a list
read_csv:
path: users.csv
register: users
delegate_to: localhost
- debug:
msg: 'User {{ users.list.1.name }} has UID {{ users.list.1.uid }} and GID {{ users.list.1.gid }}'
# Example CSV file without header and semi-colon delimiter
#
# dag;500;500
# jeroen;501;500
# Read a CSV file without headers
- name: Read users from CSV file and return a list
read_csv:
path: users.csv
fieldnames: name,uid,gid
delimiter: ';'
register: users
delegate_to: localhost
'''
RETURN = r'''
dict:
description: The CSV content as a dictionary.
returned: success
type: dict
sample:
dag:
name: dag
uid: 500
gid: 500
jeroen:
name: jeroen
uid: 501
gid: 500
list:
description: The CSV content as a list.
returned: success
type: list
sample:
- name: dag
uid: 500
gid: 500
- name: jeroen
uid: 501
gid: 500
'''
import csv
from ansible.module_utils.basic import AnsibleModule
from ansible.module_utils._text import to_text
# Add Unix dialect from Python 3
class unix_dialect(csv.Dialect):
"""Describe the usual properties of Unix-generated CSV files."""
delimiter = ','
quotechar = '"'
doublequote = True
skipinitialspace = False
lineterminator = '\n'
quoting = csv.QUOTE_ALL
csv.register_dialect("unix", unix_dialect)
def main():
module = AnsibleModule(
argument_spec=dict(
path=dict(type='path', required=True, aliases=['filename']),
dialect=dict(type='str', default='excel'),
key=dict(type='str'),
fieldnames=dict(type='list'),
unique=dict(type='bool', default=True),
delimiter=dict(type='str'),
skipinitialspace=dict(type='bool'),
strict=dict(type='bool'),
),
supports_check_mode=True,
)
path = module.params['path']
dialect = module.params['dialect']
key = module.params['key']
fieldnames = module.params['fieldnames']
unique = module.params['unique']
if dialect not in csv.list_dialects():
module.fail_json(msg="Dialect '%s' is not supported by your version of python." % dialect)
dialect_options = dict(
delimiter=module.params['delimiter'],
skipinitialspace=module.params['skipinitialspace'],
strict=module.params['strict'],
)
# Create a dictionary from only set options
dialect_params = dict((k, v) for k, v in dialect_options.items() if v is not None)
if dialect_params:
try:
csv.register_dialect('custom', dialect, **dialect_params)
except TypeError as e:
module.fail_json(msg="Unable to create custom dialect: %s" % to_text(e))
dialect = 'custom'
try:
f = open(path, 'r')
except (IOError, OSError) as e:
module.fail_json(msg="Unable to open file: %s" % to_text(e))
reader = csv.DictReader(f, fieldnames=fieldnames, dialect=dialect)
if key and key not in reader.fieldnames:
module.fail_json(msg="Key '%s' was not found in the CSV header fields: %s" % (key, ', '.join(reader.fieldnames)))
data_dict = dict()
data_list = list()
if key is None:
try:
for row in reader:
data_list.append(row)
except csv.Error as e:
module.fail_json(msg="Unable to process file: %s" % to_text(e))
else:
try:
for row in reader:
if unique and row[key] in data_dict:
module.fail_json(msg="Key '%s' is not unique for value '%s'" % (key, row[key]))
data_dict[row[key]] = row
except csv.Error as e:
module.fail_json(msg="Unable to process file: %s" % to_text(e))
module.exit_json(dict=data_dict, list=data_list)
if __name__ == '__main__':
main()

@ -0,0 +1,142 @@
# Create basic CSV file
- name: Create unique CSV file
copy:
content: |
name,uid,gid,gecos
dag,500,500,Dag Wieërs
jeroen,501,500,Jeroen Hoekx
dest: users_unique.csv
# Read a CSV file and access user 'dag'
- name: Read users from CSV file and return a dictionary
read_csv:
path: users_unique.csv
key: name
register: users_unique
- assert:
that:
- users_unique.dict.dag.name == 'dag'
- users_unique.dict.dag.gecos == 'Dag Wieërs'
- users_unique.dict.dag.uid == '500'
- users_unique.dict.dag.gid == '500'
- users_unique.dict.jeroen.name == 'jeroen'
- users_unique.dict.jeroen.gecos == 'Jeroen Hoekx'
- users_unique.dict.jeroen.uid == '501'
- users_unique.dict.jeroen.gid == '500'
# Read a CSV file and access the first item
- name: Read users from CSV file and return a list
read_csv:
path: users_unique.csv
register: users_unique
- assert:
that:
- users_unique.list.0.name == 'dag'
- users_unique.list.0.gecos == 'Dag Wieërs'
- users_unique.list.0.uid == '500'
- users_unique.list.0.gid == '500'
- users_unique.list.1.name == 'jeroen'
- users_unique.list.1.gecos == 'Jeroen Hoekx'
- users_unique.list.1.uid == '501'
- users_unique.list.1.gid == '500'
# Create basic CSV file using semi-colon
- name: Create non-unique CSV file using semi-colon
copy:
content: |
name;uid;gid;gecos
dag;500;500;Dag Wieërs
jeroen;501;500;Jeroen Hoekx
dag;502;500;Dag Wieers
dest: users_nonunique.csv
# Read a CSV file and access user 'dag'
- name: Read users from CSV file and return a dictionary
read_csv:
path: users_nonunique.csv
key: name
unique: no
delimiter: ';'
register: users_nonunique
delegate_to: localhost
- assert:
that:
- users_nonunique.dict.dag.name == 'dag'
- users_nonunique.dict.dag.gecos == 'Dag Wieers'
- users_nonunique.dict.dag.uid == '502'
- users_nonunique.dict.dag.gid == '500'
- users_nonunique.dict.jeroen.name == 'jeroen'
- users_nonunique.dict.jeroen.gecos == 'Jeroen Hoekx'
- users_nonunique.dict.jeroen.uid == '501'
- users_nonunique.dict.jeroen.gid == '500'
# Read a CSV file using an non-existing dialect
- name: Read users from CSV file and return a dictionary
read_csv:
path: users_nonunique.csv
dialect: placebo
register: users_placebo
ignore_errors: yes
- assert:
that:
- users_placebo is failed
- users_placebo.msg == "Dialect 'placebo' is not supported by your version of python."
# Create basic CSV file without header
- name: Create unique CSV file without header
copy:
content: |
dag,500,500,Dag Wieërs
jeroen,501,500,Jeroen Hoekx
dest: users_noheader.csv
delegate_to: localhost
# Read a CSV file and access user 'dag'
- name: Read users from CSV file and return a dictionary
read_csv:
path: users_noheader.csv
key: name
fieldnames: name,uid,gid,gecos
register: users_noheader
- assert:
that:
- users_noheader.dict.dag.name == 'dag'
- users_noheader.dict.dag.gecos == 'Dag Wieërs'
- users_noheader.dict.dag.uid == '500'
- users_noheader.dict.dag.gid == '500'
- users_noheader.dict.jeroen.name == 'jeroen'
- users_noheader.dict.jeroen.gecos == 'Jeroen Hoekx'
- users_noheader.dict.jeroen.uid == '501'
- users_noheader.dict.jeroen.gid == '500'
# Create broken file
- name: Create unique CSV file
copy:
content: |
name,uid,gid,gecos
dag,500,500,Dag Wieërs
jeroen,501,500,"Jeroen"Hoekx"
dest: users_broken.csv
# Read a broken CSV file using strict
- name: Read users from a broken CSV file
read_csv:
path: users_broken.csv
key: name
strict: yes
register: users_broken
ignore_errors: yes
- assert:
that:
- users_broken is failed
- "'Unable to process file' in users_broken.msg"
Loading…
Cancel
Save