Rehaul Configuration

[zaquestion]

lab currently uses hcl as it's config format, and offers limited configuration. In large part minimal config options have been added to lab because hcl hasn't played super nicely with viper overall (particularly the automatic env vars functionality, which we want to leverage). For 1.0 I'd like to completely redo this, with a simpler file format that plays nice with automatic env vars. The main issue that arises has to do with ambiguity among file formats on if keys are implicitly treated as arrays or not, hcl has this issue, I think yaml may as well. So I suspect json is going to be a top candidate here. It's also possible that viper simply has better support for this in other formats (regardless of array considerations), so it's worth looking into yaml and ini (git config like option).

We aren't setup for this today, but in theory by using viper we should be able to read config from any number of formats. So we don't need to enforce what lab accepts, we should allow any viper formats to work (where reasonable / enabling ext suffices), but we should also settle on what the default is going to be go forward.

Lastly, out of the box this should provide improved (if not complete) support for #151 (multiple gitlab providers), this includes:

• Looking for config not just in the current directory, but in the repo root (hopefully tackled in pre 1.0 as well)
• Looking up credentials based on by host
• Support reading tokens from command output (such as pass) Read token from other sources (like secret/password manager) #382

Related issues:

• config file usage? #402 -- Config Questions
• Read token from other sources (like secret/password manager) #382 -- load config from other sources
• Load Config From Env #155 -- first env var issues that arised

[prarit]

Personally I'd like to see an ini format but that's just me :). Let me look into this to see what I can come up with. To be clear, you have no objection to continuing to use viper. It's just that viper doesn't handle hcl very well, correct?

[rsteube]

Big fan of yaml personally, but toml is prefered by a lot for it's more unambiguous syntax (it's also pretty similar to ini).
I find json not so great for configs that are edited by hand.

[zaquestion]

To be clear, you have no objection to continuing to use viper. It's just that viper doesn't handle hcl very well, correct?

Correct, and more specifically, I want automatic environment vars to work more or less out of the box and not require all manner of shennanigans to work https://github.com/zaquestion/lab/blob/master/main.go#L54-L137

Had forgotten what toml looked like when I quickly wrote this up this morning. Agree toml seems like the stronger route (over ini) to supporting "git like" configs.

[prarit]

Big fan of yaml personally, but toml is prefered by a lot for it's more unambiguous syntax (it's also pretty similar to ini).
I find json not so great for configs that are edited by hand.

Ditto and I think we're all in agreement that toml looks like the best approach moving forward.

[prarit]

I would like to implement something that differentiates between a local and global config, just like git does. For example, if there is a local .git/lab/config, the settings in that file would override the ~/.config/lab/config settings. Doing that, however, increases complexity as there would have to be a 'lab config' command that had a --local and/or --global option.

An example: .git/lab/config contains
[gitlab.com]
token 12345abcde
user prarit

that overrides the ~/.config/lab/config

[gitlab.com]
token 67890vwxyz
user john.doe

The config code would always default to searching global first, and then overriding with the local values.

[prarit]

Another reason we want to have global and local configs: On some projects I'm going to want the comments always displayed when doing a 'show mr', and on other projects I'm likely not going to care unless I really want it. Having a

[mr show]
comments = true

would easily fix that type of problem.

[prarit]

A few things to report. I don't think the HCL code is working at all. viper.GetString("core.token"), for example, is always empty.

I've filed

spf13/viper#967

to see if there is any feedback. I've tried YAML (which works!), and HCL & TOML (both of which fail).

[prarit]

My plan is to:

1. Convert the existing viper infrastructure to use TOML instead of HCL. This requires a new config function ConvertHCLtoTOML that would convert existing HCL files to TOML. It would be run from main().

As noted in the previous comment, however, spf13/viper#967 is a problem. It could be that I just need a gentle tap on the head with a cluebat ;) and there is some subtle nature of using the other formats that I don't understand.

2. Update the existing viper infrastructure to always override when using environment vars LAB_CORE_HOST, LAB_CORE_TOKEN, and LAB_CORE_USER. This should be straight forward.

3. Keeping in mind Documentation for configuration #2, update the existing order of ./, ~/.config/, and .git/ to be

.git/lab
~/.config/lab
user specified config file (--configfile --conf --config?)

This config detection order would allow users to specify a per-tree host, token, and user and easily override the global ~/.config/lab entry.

While this is written focusing on the core entries (host, token, and user), it should be noted there are other config entries of tls.skip_verify and tls.ca_file. I am aware that these configs also need to be handled appropriately.

[bmeneg]

May I also suggest to move the configLoad() code into internal/config/config.go as Load() (or something similar)?
Thus we have a clear separation of configuration organization related code and its usage.

[prarit]

I have the first part of the plan above staged in #414.

[prarit]

4. Use ~/.config/lab/lab.toml <- according to a colleague it is bad practice to leave a config in the .config directory.

[prarit]

The second part is also staged in #414