TOML support

[bayerj]

Hello,

Is there a chance to support different configuration formats, e.g. TOML or JSON? I wonder how deeply yaml is integrated.

Best, -Justin

PS: Nice project. :)

[eladrich]

Hi Justin, That's a great question! The yaml integration is localized to the following parts of pyrallis

1. The pyrallis.dump function (which uses the general pyrallis.encode)
2. The pyrallis.load function (which uses the general pyrallis.decode)
3. The loading mechanism where it is used to both load the config_path and parse all cmd strings (to ensure a uniform parsing syntax as explained here)

I've been thinking about what would be a good way to support other formats as well, but decided to first release the package as it is, and maybe add support in the future.

The thing is that for 1-2 I can always just add additional functions or arguments, the confusing part is the loading mechanism. Say someone wants to pass the following value to pyrallis {1: 2}, right now pyrallis uses only yaml for parsing strings so

python train.py --int_dict="{1: 2}"

and

int_dict: {1: 2}

would function the same way and would be parsed correctly.

Now, for jsons int keys in dictionaries are not allowed, that would mean that the console mode

python train.py --int_dict="{1: 2}"

should also fail when using jsons (otherwise we are in a confusing situation where it can be loaded from cmd but not from file). So what happens here that the functionally of pyrallis changes when a different format is used, and I want to make sure that there's a way to allow that without confusing users.

I would give it some more thought and would be happy to hear if you have any specific ideas on that matter.

[eladrich]

Added an option in MR #7 to globally change the configuration format

pyrallis.set_config_type('json')
# or
pyrallis.set_config_type(pyrallis.ConfigType.JSON)

Still need to review and update docs.