arghandler

argparse extended with awesome feature enhancements to make life easier


Keywords
argparse, command-line, parsing
License
Apache-2.0
Install
pip install arghandler==1.3.1

Documentation

arghandler Build Status

Making argparse even more awesome

I love argparse, but there are some things that it simply doesn't help with as much as I'd like. Enter arghandler.

The goal behind arghandler is to provide all the capabilities of argparse plus some high-level capabilities that crop up a lot when writing command-line tools: the library aims for high quality command line interfaces with (even more) minimal code.

At present, arghandler provides two key capabilities:

  1. Adding subcommands with basically zero extra lines of code. This gives support for writing programs like git and svn which have nested subcommands.

  2. Configuring the logging framework (e.g., the desired logging level) from the command line - again with basically one line of code.

We have lots more improvements we want to add - and as we have time and receive feedback, we'll add more features.

If you have ideas, email me or code it up and generate a pull request!

Installation

Use pip or easy_install to install the library:

pip install arghandler

or

easy_install arghandler

You can find arghandler on pypi for relevant details should you need them.

Usage

Just like with argparse.ArgumentParser, in arghandler everything revolves around ArgumentHandler. In fact, it's (not so secretly) a subclass of ArgumentParser, so you can use it exactly the way you use ArgumentParser. But ArgumentHandler has some new tricks.

To benefit from ArgumentHandler, your command-line configuration code will follow this logic:

from arghandler import ArgumentHandler

handler = ArgumentHandler() # this accepts all args supported by ArgumentParser

# config the handler using add_argument, set_logging_level, set_subcommands, etc...

handler.run() # throw the configured handler at an argument string!

Now for some details...

Invoking ArgumentHandler

ArgumentHandler can be invoked on arguments in two ways.

ArgumentHandler.parse_args([argv]) is little different from ArgumentParser.parse_args([argv]). If argv is omitted, then the value of sys.argv is used. The only notable differences are:

  • If a logging argument was set, then this will be included in the namespace object returned.

  • If subcommands are available, then the subcommand will be given by the value of args.cmd and the subcommand's arguments will be given by args.cargs.

ArgumentHandler.run(argv,context_fxn) makes the class perform its more unique and powerful capabilities. Notably: configuring the logger and running subcommands. As with parse_args(...), if argv is not specified, then sys.argv will be used. The context_fxn is also optional and is used as part of subcommand processing. See that section below for more details.

Enabling autocompletion

When constructing an ArgumentHandler, you can enable autocompletion. This requires doing two separate things.

First, pass the keyword argument enable_autocompetion=True to ArgumentHandler(...).

Second, in the top-level script that will be your command-line tool, include the line

# PYTHON_ARGCOMPLETE_OK

near the top (in the first 1024 bytes). For more details on this, see the argcomplete documentation.

For an example of this in action, see examples/dummy.py!.

Setting the logging level

If you use the python logging package, this feature will save you some time.

The ArgumentParser.set_logging_argument(...) method allows you to specify a command-line argument that will set the logging level. The method accepts several arguments:

ArgumentParser.set_logging_argument(*names,default_level=logging.ERROR,config_fxn=None)
  • *names stands in for one or more arguments that specify the argument names that will be used. These follow the same rules as ones passed into ArgumentParser.add_argument(...). Moreover, they MUST be optional arguments (i.e., start with a '-' character).

  • default_level indicates the default level the logging framework will be set to should the level not be specified on the command line.

  • config_fxn allows the developer to write special logging configuration code. If not specified, the logging.basicConfig function will be invoked with the appropriate logging level. The function must accept two arguments: the logging level and the namespace args object returned by the ArgumentParser.parse_args method. The configuration itself will happen when the ArgumentHandler.run(...) method is called.

If you're cool with the defaults in basicConfig, then your method call will look something like this

handler.set_logging_argument('-l','-log_level',default_level=logging.INFO)

If you do want to do some customization, then your code will look like this

handler.set_logging_argument('-l','-llevel',
	config_fxn=lambda level,args: logging.basicConfig(level=level,format='%(message)'))

Declaring subcommands using decorators

This feature makes it possible to write nested commands like git commit and svn checkout with basically zero boilerplate code. To do this arghandler provides the @subcmd decorator. To declare a subcommand, just put the decorator on the function you want to act as the subcommand.

from arghandler import *

@subcmd
def echo(parser,context,args):
	print ' '.join(args)

# here we associate the subcommand 'foobar' with function cmd_foobar
@subcmd('foobar', help = 'Does foobar')
def cmd_foobar(parser,context,args):
	print 'foobar'

handler = ArgumentHandler()
handler.run(['echo','hello','world']) # echo will be called and 'hello world' will be printed

Notice that the subcommands always take three arguments.

args is the set of arguments that follow the subcommand on the command line.

context is an object that can make valuable global information available to subcommands. By default, the context is the namespace object returned by the internal call to ArgumentHandler.parse_args(...). Other contexts can be produced by passing a context-producing function to the ArgumentHandler.run(...) function:

@subcmd('ping')
def ping_server(parser,server_address,args):
	os.system('ping %s' % server_address)

handler = ArgumentHandler()
handler.add_argument('-s','--server')

# when this is run, the context will be set to the return value of context_fxn
# in this case, it will be the string '127.0.0.1'
handler.run(['-s','127.0.0.1','ping'],context_fxn=lambda args: args.server

Finally, parser is an instance of argparse.ArgumentParser which has been preconfigured to behave properly for the subcommand. Most crucially, this means that parser.prog is set to <top_level_program> <sub_command> so that help messages print out correctly for the subcommand. Should your subcommand want to parse arguments, this parser object should be used.

Declaring subcommands without decorators

While decorators are the preferred way to specify subcommands, subcommands can also be specified using the ArgumentHandler.set_subcommands(...) function. This method expects a dictionary: keys are command names, values are the command functions:

from arghandler import *

def echo(parser,context,args):
	print ' '.join(args)

def cmd_foobar(parser,context,args):
	print 'foobar'

handler = ArgumentHandler()
handler.set_subcommands( {'echo':echo, 'foobar':cmd_foobar} )
handler.run(['echo','hello','world']) # echo will be called and 'hello world' will be printed

All the logic and rules around the context function apply here. Moreoever, the complete set of subcommands include those specified using decorators AND those specified through the set_subcommands(...) method.

Making subcommands in subcommands

One valuable use for the set_subcommands(...) method is implementing subcommand options for a subcommand. For example, suppose you want a program with the following command subtree:

power
  - create
    - config
    - proj
  - run
    - all
    - proj

In this case, create and run would be top-level subcommands that could be declared using standard subcmd decorators. But what about the config and proj commands underneath create? These can be created using a new ArgumentHandler inside the create function like this:

def create_config(parser, context, args):
    parser.add_argument('location')
    args = parser.parse_args(args)

    # do stuff

    return

def create_proj(parser, context, args):
    parser.add_argument('name')
    args = parser.parse_args(args)

    print(f'Creating the project: {args.name}')

    # do stuff

    return


@subcmd('create', help='create a resource')
def create(parser, context, args):
    handler = ArgumentHandler()

    handler.set_subcommands({'config': (create_config, 'create a config file'),
                             'proj': (create_proj, 'create a project')
                            },
                            use_registered_subcmds=False)
    
    handler.run(args)

Note the use of use_registered_subcmds=False - this is important to omit any functions globally registered as commands using the @subcmd decorator.

Setting the help message

The format of the help message can be set to one more friendly for subcommands by passing the ArgumentHandler constructor the keyword argument use_subcommand_help=True.

This will produce a help message that looks something like this:

usage: test.py [-h] subcommand

positional arguments:
  subcommand
    cmd1  cmd1_help_str

optional arguments:
  -h, --help  show this help message and exit

Some best practices

Use ArgumentParser or ArgumentHandler inside subcommands. This will ensure that informative help messages are available for all your subcommands.

from arghandler import *

@subcmd
def echo(parser,context,args):
	parser.add_argument('-q','--quote_char',required=True)
	args = parser.parse_args(args)
	print '%s%s%s' % (args.quote_char,' '.join(args),args.quote_char)

@subcmd('foobar')
def cmd_foobar(parser,context,args):
	print 'foobar'

handler = ArgumentHandler()
handler.run(['echo','-h']) # the help message for echo will be printed

Use logging. Logging gives you much more control over what debugging/informational content is printed out by your program. And with arghandler it's easier than ever to configure from the command line!