Why doesn't astropy use ~/.config/astropy by default (on Unix-like systems)?

[weaverba137]

If $XDG_CONFIG_HOME is set, astropy will look for configuration files in $XDG_CONFIG_HOME/astropy/. However, it will not look for configuration files in $XDG_CONFIG_HOME/astropy/config/. Similarly, if $XDG_CACHE_HOME is set, astropy will look for cache files in $XDG_CACHE_HOME/astropy/, not $XDG_CACHE_HOME/astropy/cache. Why does the layout of $XDG_CONFIG_HOME/astropy/ have to be different from the layout of $HOME/.astropy/, which contains config/ and cache/ directories? Unless one knows that the layout is different, one would expect that $XDG_CONFIG_HOME/astropy/config/ and $XDG_CACHE_HOME/astropy/cache/ would be the directories to contain configuration and cache files.

[saimn]

Because XDG_CONFIG_HOME and XDG_CACHE_HOME are Linux standards for config and cache (usually ~/.config and ~/.cache). For me the current behavior makes perfect sense, if XDG_CONFIG_HOME is set it goes to ~/.config/astropy (typically**) otherwise it is ~/.astropy/config (and the same for cache).

** Actually if XDG_CONFIG_HOME is not set, it should default to $HOME/.config, so Astropy uses this in a diverted way, which is confusing, see also #6516 recently. Maybe it would have been better to have an ASTROPY_CONFIG_DIR variable instead to customize the $HOME/.astropy/ paths directly, instead of trying to follow the xdg spec.

[weaverba137]

OK, maybe I should rephrase this: Why doesn't astropy use ~/.config/astropy by default (on Unix-like systems)?

[Cadair]

@weaverba137 I came here because I wanted to know the answer to "Why doesn't astropy use ~/.config/astropy by default (on Unix-like systems)?" Should we make this the title of the issue or something?

[weaverba137]

I would be happy to change the title.

[Cadair]

So from get_config_dir docstring:

This directory is typically $HOME/.astropy/config, but if the XDG_CONFIG_HOME environment variable is set and the $XDG_CONFIG_HOME/astropy directory exists, it will be that directory. If neither exists, the former will be created and symlinked to the latter.

This is not the "correct" behaviour on a linux system as far as I am aware.

[Cadair]

ping @eteq as he wrote it apparently.

[pllim]

I would say that this ship has sailed. Unless we do something drastic like #8245, too much overhead in refactoring (including answering issues from confused users on the behavior change) for little reward.

[weaverba137]

OK, but as @Cadair said:

This is not the "correct" behaviour on a linux system as far as I am aware.

Can we at least document that we are aware that this is not correct behavior?

[Cadair]

We should absolutely do #8245 and I am -1 on closing this as wont-fix.

[pllim]

Feel free to open a PR for the desired documentation, as you already have an idea anyway.

As for #8245, it is not trivial. Maybe for 5.0 if someone is willing to invest the time and energy, but not before.

[Cadair]

I agree it's not easy, but this bug should be fixed which is why I don't want to close it. 😄