list using alternative to multipath about platformdirs HOT 8 CLOSED

this would be a pain to support for a Path Type

We could introduce the contract that path variants always return the first of the multipath. If users want multipath they would need to do: user_data_dir().split(os.pathsep), at which point they can cast the elements into Path if they'd wish to 🤔

from platformdirs.

We could introduce the contract that path variants always return the first of the multipath.

Alternatively, site_data_path could return a list of pathlib.Path objects if multipath=True was passed. Ideally, site_data_dir would return a list of strings in that case, too (as suggested by @RonnyPfannschmidt). But this wouldn't be backwards compatible, correct?

From what I can see, site_data_dir is the only affected API here, correct?

from platformdirs.

It also would mutate the type returned based on a boolean flag which I don't like at all 😬

from platformdirs.

We could introduce the contract that path variants always return the first of the multipath.

In other words, the multipath flag would have no effect on site_data_path. It will simply always return the first path, which is the default for site_data_dir if multipath is not set.

How come that there is a binary flag for this anyway? Wouldn't it make more sense to have two new api calls: site_data_dir_list and site_data_path_list?

from platformdirs.

Because for dirs the return type does not change 😬 because we don't return the paths as a list are but joined by os.pathsep. obviously we can't do that for pathlib.

from platformdirs.

Because for dirs the return type does not change

That is of course correct, but their output changes semantically, i.e. as a user I have to handle multipath output differently than singlepath output. It looks to me like all the arguments of a boolean-flag trap apply here as well. Nonetheless, #27 follows your suggestion to only return the first item in case site_data_dir is called with multipath=True.

from platformdirs.

That is of course correct, but their output changes semantically, i.e. as a user I have to handle multipath output differently than singlepath output.

Not really, UNIX site config values always return paths joined by os.pathsep... In simple case just happens to not matter, because joining an array of length 1 is always the same as the array element.

from platformdirs.

We solved this for now by picking first for paths. str endpoints still work.

from platformdirs.