add get_bool_env for parsing bool-like env vars

[IanButterworth]

Expands and brings @giordano's get_bool_env over from Pkg.

For another PR: Use where appropriate in Base & stdlibs

[StefanKarpinski]

Obvious next move after this is merged is to delete the code from Pkg and use this 😃

[giordano]

Looks familiar 😄 But I appreciate the more robust error handling than what's in Pkg now

[rfourquet]

What about exporting this?

[DilumAluthge]

It's really just for internal use, right? Also, I can imagine wanting to make breaking changes to it in the future.

[giordano]

It's really just for internal use, right?

Well, it's in the manual at the moment

[DilumAluthge]

Oh I missed that change. Personally, I would remove it from the manual, but I don't feel particularly strongly about it.

[ericphanson]

as someone with various less-good versions of this in different places, I think this would be pretty useful to have outside of Base!

[nickrobinson251]

probably this doesn't matter, as this won't be called in a performance critical place, but... maybe we should avoid this allocating when possible, by checking if the string is already lowercase (which i think should be okay overhead in practice since we're not expecting really long strings here)

Suggested change

	val = lowercase(get(ENV, name, ""))
	val = get(ENV, name, "")
	val = all(islowercase, val) ? val : lowercase(val)

idk, just an thought

[nickrobinson251]

oh on the other hand this is a whole can of worms so maybe ignore this if all(islowercase, str) is controversial
#46444

(i just want less stuff to allocate 😛)

[rfourquet]

If calling lowercase is worrying, adding the explicit list in truthy is also an option, at least if restricted to all lower or upper cases (i.e. not including "TrUe"...).

[IanButterworth]

Updated with the truthy and falsy constructors expanded and moved out as consts

[KristofferC]

Imo, a function that deals with external data like this shouldn't throw. It should instead return something to the user indicating the error.

Otherwise, you are always going to have to wrap usage of this function in a try catch since you cannot predict what is in the env var, and you are unlikely to want to panic just due to a malformed env variable.

[DilumAluthge]

Seems like returning nothing would make sense here?

[IanButterworth]

Sounds good. I just switched it over to this

[KristofferC]

This feels pretty ad hoc to me. Like how the words for accepting true/false are in english and the exact form of the words that are accepted. To me at least it doesn't immediately feel like something that should be made a part of the language itself.

[giordano]

How about not making this public API (i.e. removing its mention from NEWS.md and docs)? At least it can be used internally in Julia to make parsing of the environment variables more consistent. Defining a better API can be done at a later stage, if one wants to push for it.

[IanButterworth]

How about not making this public API (i.e. removing its mention from NEWS.md and docs)?

Yeah, sounds good. Done

[IanButterworth]

Does this look good to go?

julia> @time Base.get_bool_env("FOO", true)
  0.000002 seconds
true

julia> ENV["FOO"] = ""
""

julia> @time Base.get_bool_env("FOO", true)
  0.000007 seconds
true

julia> ENV["FOO"] = "0"
"0"

julia> @time Base.get_bool_env("FOO", true)
  0.000007 seconds (3 allocations: 88 bytes)
false

julia> ENV["FOO"] = "1"
"1"

julia> @time Base.get_bool_env("FOO", true)
  0.000008 seconds (2 allocations: 56 bytes)
true

julia> 

julia> ENV["FOO"] = "11"
"11"

julia> @time Base.get_bool_env("FOO", true)
  0.000007 seconds (3 allocations: 88 bytes)

julia> 

[DilumAluthge]

Since it's not public, we can always tweak it further in the future.