The argparse_config utility reads defaults for commandline args from a config file. The cute thing is, it figures out what config options to expect based on your argparse commandline args definition.
Let's say I'm reimplementing the Mercurial commandline client. I specify the commandline argument processing with argparse, of course:
>>> arg_parser = ArgumentParser('hg') >>> arg_parser.add_argument('--repository') >>> sub_parsers = arg_parser.add_subparsers() >>> >>> merge_parser = sub_parsers.add_parser('merge') >>> merge_parser.add_argument('--tool') >>> merge_parser.add_argument('--force', action='store_true', default=False) >>> >>> commit_parser = sub_parsers.add_parser('commit') >>> commit_parser.add_argument('--user') >>> commit_parser.add_argument('--message')
When I go to use this client, though, I have to keep specifying my --user with every commit, and --tool with every merge. That sucks! What I want is to have my client understand a simple config file format:
[merge] tool: meld [commit] user: Tikitu de Jager <email@example.com>
And obviously, as I add more arguments and subcommands to my client, it should allow me to add defaults in the config file without writing more code.
This is what argparse_config gives you. To use it with the mercurial client arg_parser above:
>>> import argparse_config >>> argparse_config.read_config(arg_parser, '/home/tikitu/.my_hg.cfg')
... and that's it. Calling arg_parser.parse_args() will parse args as usual, but the default values will be taken from the config file, if they are given there:
>>> parsed_args = arg_parser.parse_args(['merge']) >>> parsed_args.tool 'meld'
What can I put in the config file?
Under the hood argparse_config uses the standard library ConfigParser. Arguments that aren't for a subcommand go in the section [default]. The names are munged from the commandline argument, removing leading dashes and converting internal dashes to underscores (e.g. --log-level becomes log_level:).
Flags (i.e. commandline args that take no parameters) are turned on if present in the config, just like the commandline:
is the equivalent of --verbose. Either verbose: or verbose will work, but (watch out!) verbose: a-value doesn't do anything different to verbose.
Writing a config file from some commandline arguments
Included in the package is a utility to generate a config file following these rules, from a given set of commandline arguments. That looks like this:
>>> parsed_args = arg_parser.parse_args(['--repository', 'https://bitbucket.org/tikitu/argparse_config', 'merge']) >>> print argparse_config.generate_config(arg_parser, parsed_args, only_non_defaults=True) [default] repository: https://bitbucket.org/tikitu/argparse_config [merge] [commit]
Some complications make this less useful than it could be, sadly:
- If you use subcommands, you can only parse the args for one of them at a time.
- We can't tell the difference between default values written in code (which should not be added to the config file) and written in a previously-read-in config file (which should). This is why only_non_defaults exists.
How does it work?
By gudgeling about in the private internals of argparse. Yes, that's not pretty.
Any required arguments that are present in a config file will show as optional, not required, in the --help output. (This is a bug-by-design, due to not having any clever idea about how to do it better.) It may help to tell yourself, "It's not required on the commandline because I gave it in the config file." (I will gladly make this dodgy rationalisation disappear if I figure out how to handle required arguments more tidily.)
It's on BitBucket. Feel free to play. It comes with a handy zc.buildout wrapper too, overkill though that clearly is.
It's "alpha software" at present; likely to be buggy and lots of stuff ain't there yet. Check the issues list to stay up to date. Some things I plan to add:
- A utility to output a config file, based on a set of commandline arguments.