ConfigArgParse
Overview
Applications with more than a handful of user-settable options are best configured through a combination of command line args, config files, hard-coded defaults, and in some cases, environment variables.
Python's command line parsing modules such as argparse have very limited support for config files and environment variables, so this module extends argparse to add these features.
Available on PyPI: http://pypi.python.org/pypi/ConfigArgParse
Features
- command-line, config file, env var, and default settings can now be defined, documented, and parsed in one go using a single API (if a value is specified in more than one way then: command line > environment variables > config file values > defaults)
- config files can have .ini or .yaml style syntax (eg. key=value or key: value)
- user can provide a config file via a normal-looking command line arg (eg. -c path/to/config.txt) rather than the argparse-style @config.txt
- one or more default config file paths can be specified (eg. ['/etc/bla.conf', '~/.my_config'] )
- all argparse functionality is fully supported, so this module can serve as a drop-in replacement (verified by argparse unittests).
- env vars and config file keys & syntax are automatically documented in the -h help message
- new method
print_values()can report keys & values and where they were set (eg. command line, env var, config file, or default). - lite-weight (no 3rd-party library dependencies except (optionally) PyYAML)
- extensible (
ConfigFileParsercan be subclassed to define a new config file format) - unittested by running the unittests that came with argparse but on configargparse, and using tox to test with Python 2.7 and Python 3+
Example
my_script.py:
Script that defines 4 options and a positional arg and then parses and prints the values. Also, it prints out the help message as well as the string produced by format_values() to show what they look like.
import configargparse
p = configargparse.ArgParser(default_config_files=['/etc/app/conf.d/*.conf', '~/.my_settings'])
p.add('-c', '--my-config', required=True, is_config_file=True, help='config file path')
p.add('--genome', required=True, help='path to genome file') # this option can be set in a config file because it starts with '--'
p.add('-v', help='verbose', action='store_true')
p.add('-d', '--dbsnp', help='known variants .vcf', env_var='DBSNP_PATH') # this option can be set in a config file because it starts with '--'
p.add('vcf', nargs='+', help='variant file(s)')
options = p.parse_args()
print(options)
print("----------")
print(p.format_help())
print("----------")
print(p.format_values()) # useful for logging where different settings came from
config.txt:
Since the script above set the config file as required=True, lets create a config file to give it:
# settings for my_script.py
genome = HCMV # cytomegalovirus genome
dbsnp = /data/dbsnp/variants.vcf
command line:
Now run the script and pass it the config file:
DBSNP_PATH=/data/dbsnp/variants_v2.vcf python config_test.py --my-config config.txt f1.vcf f2.vcf
output:
Here is the result:
Namespace(dbsnp='/data/dbsnp/variants_v2.vcf', genome='HCMV', my_config='config.txt', v=False, vcf=['f1.vcf', 'f2.vcf'])
----------
usage: config_test.py [-h] -c MY_CONFIG --genome GENOME [-v] [-d DBSNP]
vcf [vcf ...]
Args that start with '--' (eg. --genome) can also be set in a config file
(/etc/app/conf.d/*.conf or ~/.my_settings or specified via -c). Config file
syntax allows: key=value, flag=true, stuff=[a,b,c] (for details, see syntax at
https://goo.gl/R74nmi). If an arg is specified in more than one place, then
commandline values override environment variables which override config file
values which override defaults.
positional arguments:
vcf variant file(s)
optional arguments:
-h, --help show this help message and exit
-c MY_CONFIG, --my-config MY_CONFIG
config file path
--genome GENOME path to genome file
-v verbose
-d DBSNP, --dbsnp DBSNP
known variants .vcf [env var: DBSNP_PATH]
----------
Command Line Args: --my-config config.txt f1.vcf f2.vcf
Environment Variables:
DBSNP_PATH: /data/dbsnp/variants_v2.vcf
Config File (config.txt):
genome: HCMV
Special Values
Under the hood, configargparse handles environment variables and config file values by converting them to their corresponding command line arg. For example, "key = value" will be processed as if "--key value" was specified on the command line.
Also, the following special values (whether in a config file or an environment variable) are handled in a special way to support booleans and lists:
key = trueis handled as if "--key" was specified on the command line. In your python code this key must be defined as a boolean flag (eg. action="store_true" or similar).key = [value1, value2, ...]is handled as if "--key value1 --key value2" etc. was specified on the command line. In your python code this key must be defined as a list (eg. action="append").
Config File Syntax
Only command line args that have a long version (eg. one that starts with '--') can be set in a config file. For example, "--color" can be set by putting "color=green" in a config file. The config file syntax depends on the constuctor arg: config_file_parser_class which can be set to one of the provided classes: DefaultConfigFileParser, YAMLConfigFileParser, ConfigparserConfigFileParser or to your own subclass of the ConfigFileParser abstract class.
DefaultConfigFileParser - the full range of valid syntax is:
# this is a comment
; this is also a comment (.ini style)
--- # lines that start with --- are ignored (yaml style)
-------------------
[section] # .ini-style section names are treated as comments
# how to specify a key-value pair (all of these are equivalent):
name value # key is case sensitive: "Name" isn't "name"
name = value # (.ini style) (white space is ignored, so name = value same as name=value)
name: value # (yaml style)
--name value # (argparse style)
# how to set a flag arg (eg. arg which has action="store_true")
--name
name
name = True # "True" and "true" are the same
# how to specify a list arg (eg. arg which has action="append")
fruit = [apple, orange, lemon]
indexes = [1, 12, 35 , 40]
YAMLConfigFileParser - allows a subset of YAML syntax (http://goo.gl/VgT2DU)
# a comment
name1: value
name2: true # "True" and "true" are the same
fruit: [apple, orange, lemon]
indexes: [1, 12, 35, 40]
ConfigparserConfigFileParser - allows a subset of python's configparser module syntax (https://docs.python.org/3.7/library/configparser.html). In particular the following configparser options are set:
config = configparser.ConfigParser(
delimiters=("=",":"),
allow_no_value=False,
comment_prefixes=("#",";"),
inline_comment_prefixes=("#",";"),
strict=True,
empty_lines_in_values=False,
)
Once configparser parses the config file all section names are removed, thus all keys must have unique names regardless of which INI section they are defined under. Also, any keys which have python list syntax are converted to lists by evaluating them as python code using ast.literal_eval (https://docs.python.org/3/library/ast.html#ast.literal_eval). To facilitate this all multi-line values are converted to single-line values. Thus multi-line string values will have all new-lines converted to spaces. Note, since key-value pairs that have python dictionary syntax are saved as single-line strings, even if formatted across multiple lines in the config file, dictionaries can be read in and converted to valid python dictionaries with PyYAML's safe_load. Example given below:
# inside your config file (e.g. config.ini)
[section1] # INI sections treated as comments
system1_settings: { # start of multi-line dictionary
'a':True,
'b':[2, 4, 8, 16],
'c':{'start':0, 'stop':1000},
'd':'experiment 32 testing simulation with parameter a on'
} # end of multi-line dictionary value
.......
# in your configargparse setup
import configargparse
import yaml
parser = configargparse.ConfigParser(
config_file_parser_class=configargparse.ConfigparserConfigFileParser
)
parser.add_argument('--system1_settings', type=yaml.safe_load)
args = parser.parse_args() # now args.system1 is a valid python dict
ArgParser Singletons
To make it easier to configure different modules in an application, configargparse provides globally-available ArgumentParser instances via configargparse.get_argument_parser('name') (similar to logging.getLogger('name')).
Here is an example of an application with a utils module that also defines and retrieves its own command-line args.
main.py
import configargparse
import utils
p = configargparse.get_argument_parser()
p.add_argument("-x", help="Main module setting")
p.add_argument("--m-setting", help="Main module setting")
options = p.parse_known_args() # using p.parse_args() here may raise errors.
utils.py
import configargparse
p = configargparse.get_argument_parser()
p.add_argument("--utils-setting", help="Config-file-settable option for utils")
if __name__ == "__main__":
options = p.parse_known_args()
Help Formatters
ArgumentDefaultsRawHelpFormatter is a new HelpFormatter that both adds default values AND disables line-wrapping. It can be passed to the constructor: ArgParser(.., formatter_class=ArgumentDefaultsRawHelpFormatter)
Aliases
The configargparse.ArgumentParser API inherits its class and method names from argparse and also provides the following shorter names for convenience:
- p = configargparse.get_arg_parser() # get global singleton instance
- p = configargparse.get_parser()
- p = configargparse.ArgParser() # create a new instance
- p = configargparse.Parser()
- p.add_arg(..)
- p.add(..)
- options = p.parse(..)
HelpFormatters:
- RawFormatter = RawDescriptionHelpFormatter
- DefaultsFormatter = ArgumentDefaultsHelpFormatter
- DefaultsRawFormatter = ArgumentDefaultsRawHelpFormatter
Design Notes
Unit tests:
tests/test_configargparse.py contains custom unittests for features specific to this module (such as config file and env-var support), as well as a hook to load and run argparse unittests (see the built-in test.test_argparse module) but on configargparse in place of argparse. This ensures that configargparse will work as a drop in replacement for argparse in all usecases.
Previously existing modules (PyPI search keywords: config argparse):
- argparse (built-in module Python v2.7+)
- Good:
- fully featured command line parsing
- can read args from files using an easy to understand mechanism
- Bad:
- syntax for specifying config file path is unusual (eg. @file.txt)and not described in the user help message.
- default config file syntax doesn't support comments and is unintuitive (eg. --namevalue)
- no support for environment variables
- Good:
- ConfArgParse v1.0.15 (https://pypi.python.org/pypi/ConfArgParse)
- Good:
- extends argparse with support for config files parsed by ConfigParser
- clear documentation in README
- Bad:
- config file values are processed using ArgumentParser.set_defaults(..) which means "required" and "choices" are not handled as expected. For example, if you specify a required value in a config file, you still have to specify it again on the command line.
- doesn't work with Python 3 yet
- no unit tests, code not well documented
- Good:
- appsettings v0.5 (https://pypi.python.org/pypi/appsettings)
- Good:
- supports config file (yaml format) and env_var parsing
- supports config-file-only setting for specifying lists and dicts
- Bad:
- passes in config file and env settings via parse_args namespace param
- tests not finished and don't work with Python 3 (import StringIO)
- Good:
- argparse_config v0.5.1 (https://pypi.python.org/pypi/argparse_config)
- Good:
- similar features to ConfArgParse v1.0.15
- Bad:
- doesn't work with Python 3 (error during pip install)
- Good:
- yconf v0.3.2 - (https://pypi.python.org/pypi/yconf) - features and interface not that great
- hieropt v0.3 - (https://pypi.python.org/pypi/hieropt) - doesn't appear to be maintained, couldn't find documentation
- configurati v0.2.3 - (https://pypi.python.org/pypi/configurati)
- Good:
- JSON, YAML, or Python configuration files
- handles rich data structures such as dictionaries
- can group configuration names into sections (like .ini files)
- Bad:
- doesn't work with Python 3
- 2+ years since last release to PyPI
- apparently unmaintained
- Good:
Design choices:
- all options must be settable via command line. Having options that can only be set using config files or env. vars adds complexity to the API, and is not a useful enough feature since the developer can split up options into sections and call a section "config file keys", with command line args that are just "--" plus the config key.
- config file and env. var settings should be processed by appending them to the command line (another benefit of #1). This is an easy-to-implement solution and implicitly takes care of checking that all "required" args are provied, etc., plus the behavior should be easy for users to understand.
- configargparse shouldn't override argparse's convert_arg_line_to_args method so that all argparse unit tests can be run on configargparse.
- in terms of what to allow for config file keys, the "dest" value of an option can't serve as a valid config key because many options can have the same dest. Instead, since multiple options can't use the same long arg (eg. "--long-arg-x"), let the config key be either "--long-arg-x" or "long-arg-x". This means the developer can allow only a subset of the command-line args to be specified via config file (eg. short args like -x would be excluded). Also, that way config keys are automatically documented whenever the command line args are documented in the help message.
- don't force users to put config file settings in the right .ini [sections]. This doesn't have a clear benefit since all options are command-line settable, and so have a globally unique key anyway. Enforcing sections just makes things harder for the user and adds complexity to the implementation.
- if necessary, config-file-only args can be added later by implementing a separate add method and using the namespace arg as in appsettings_v0.5
Relevant sites:
- http://stackoverflow.com/questions/6133517/parse-config-file-environment-and-command-line-arguments-to-get-a-single-coll
- http://tricksntweaks.blogspot.com/2013_05_01_archive.html
- http://www.youtube.com/watch?v=vvCwqHgZJc8#t=35
Versioning
This software follows Semantic Versioning
82 Dec 28, 2022
7 Nov 17, 2022
3.2k Jan 9, 2023
41.4k Jan 2, 2023
1.1k Jan 8, 2023
1 Aug 14, 2022
2 Dec 22, 2022
2 Dec 26, 2021
184 Jan 7, 2023
12 May 7, 2022
1 Dec 1, 2022
13 Dec 26, 2022
2.7k Jan 7, 2023
490 Dec 13, 2022
7 Dec 14, 2022
4.7k Dec 31, 2022
1k Dec 24, 2022
1.3k Dec 15, 2022