Pull request for #61: restructure YAML in template

[Shun-Yang]

@stanfordquan , please test it and check if I have achieved our goals listed in #61. Note that you will need to wait for gslab-econ/gslab-python#106 to pull in and switch to the 4.1.0 dev verson of gslab_python.

[yuchuan2016]

Excuse me for my interrupting, but do we have any rule of using 'abc' versus "abc" in a python script? I think we might want to use quotes consistently .

[qlquanle]

@arosenbe and I had a decent debate about it and we settled on consistency within scripts. I'm fine with specifying one over the other. My habit of using double quotes came from Stata.

[Shun-Yang]

Good point. I wasn't paying attention on this. All double quotes work for me.

[yuchuan2016]

I think we are using single quotes throughout the current SConscript and SConstruct, as well as all gslab_python scripts. So I would propose we stick to single quotes as default.

[yuchuan2016]

@stanfordquan , I can review this if you have not started.

[yuchuan2016]

@Shun-Yang , actually I'm not quite sure about why we need a config_user_template.yaml.

• We would prompt to users if no config_user.yaml is detected in the load_yaml_value function here and create a config_user.yaml automatically.
• We would provide default stata executable in Windows/Unix when no stata executable is specified.
• The cache path and release path in the template do not work almost 100%.
• If the user does not need stata/release/cache, why do we force them to have such a config_user.yaml file to confuse them?

Perhaps I miss something, but I do not see/remember any benefits of this template. Could you remind me of the discussion about this? I would prefer that a config_user.yaml is created only if people call it using load_yaml_value function.

[qlquanle]

Good point. The default values and prompts given by our yaml reader does serve as a "template". Once @Shun-Yang weighed in we may want to ping PIs about this.

[Shun-Yang]

@yuchuan2016 , this is indeed good point. I think the config_user_template idea is first raised here by Matt. In retrospect, I agree with you that there is really no need on creating such a template as config_user will be created anyway if the yaml reader cannot find one. Currently, we only ask users to type in their Stata executable. Do we want to also prompt for cache_dir and release_dir (of course, they can type none for all these)? This way, the user at least know that they can set these up via config_user.

[yuchuan2016]

@Shun-Yang , I think it will prompt to user to enter the release directory if people want to make a release here, and it will prompt to users to enter the cache directory if people specify mode=cache here.

This question is whether we want the repo to contain a user-config.yaml at the beginning regardless of whether it's needed or not, or create such a file when the users need one. I would vote for the second. The first is more likely to cause future breakdown, since after we copy the template, it won't prompt to users to enter the value, but just use the incorrect default value in the template.

[Shun-Yang]

Per discussion, we decide that we don't implement this config_user_template.yaml for now, as if the user does not have a config_user.yaml file, the default is to create a new one with prompt asking user for information. I confirm that both stata-executable and cache_dir prompt will properly appear. release.py is currently down. A new issue gslab-econ/gslab_python#111 will fix that.

@gentzkow FYI. Please see discussion in this thread. Let us know if you think there is still need to create config_user_template.yaml.

[Shun-Yang]

@yuchuan2016 , I have removed config_user_template. Let me know if you have other comments on this, or I can pull this in.

[yuchuan2016]

@Shun-Yang , great. I have nothing to add. We perhaps want to update the README.md to give instructions on how to turn on the dependencies. But it's up to you whether we want to update it now or update after everything has finished.