Running `parse` should not mutate the environment

[pbrisbin]

Sample code:

loadConfig :: IO String
loadConfig = Env.parse id $ Env.var Env.str "STR" mempty

main :: IO ()
main = do
    print =<< loadConfig
    print =<< loadConfig

Expected outcome

% stack build && STR=str stack exec example
"str"
"str"

Actual outcome:

% stack build && STR=str stack exec example
"str"
Available environment variables:

  STR

Parsing errors:

  STR is unset

This is breaking things for me because my app loads configuration from the environment at startup, and we load a new App value for each of our test cases. So the first test case reads the environment, unsets everything, then subsequent test cases fail.

[pbrisbin]

It looks like there is a keep modifier to prevent this, so it seems intentional. Would you mind explaining the value of having things unset by default? I'm considering setting keep all the time as a matter of course, but I wonder if I'm missing some landmine that'll bite us later, which has led to your choice in default.

[pbrisbin]

Aha, I just found #6!

As for the default behavior, although it's safer to unset everything and ask the user to mark the variables they want to keep, it's unclear to me how surprising that is.

Clearly, I was indeed surprised by this :) Would you accept a PR that reversed that default, and replaced keep with something like clear or sensitive to make the unset opt-in?

[supki]

I've come to agree that this was the wrong default, even though it'd been moderately useful in some projects with less experienced developers at $WORK. Most people don't really expect parsing to modify their environment.