ignore tree hashes on Windows w/o symlink capability

[StefanKarpinski]

Step one of addressing #3643. I implemented this by providing a default for the ignore_hash value in download_artifact, namely the default value is false unless the OS is Windows and the user cannot create symlinks in the artifact directory, in which case we default ignore_hash to true instead. This still emits a warning, which seems desirable based on the discussion in #3643.

[IanButterworth]

I think this does what you want?

Suggested change

	ignore_hash = if get(ENV, "JULIA_PKG_IGNORE_HASHES", "") == ""
	# default: false except Windows users who can't symlink
	Sys.iswindows() && !mktempdir(can_symlink, dirname(src))
	else
	Base.get_bool_env("JULIA_PKG_IGNORE_HASHES", ignore_hash)
	ignore_hash === nothing && @error(
	"Invalid ENV[\"JULIA_PKG_IGNORE_HASHES\"] value",
	ENV["JULIA_PKG_IGNORE_HASHES"],
	)
	ignore_hash = something(ignore_hash, false)
	end
	# default: false except Windows users who can't symlink
	default = Sys.iswindows() && !mktempdir(can_symlink, dirname(src))
	ignore_hash = Base.get_bool_env("JULIA_PKG_IGNORE_HASHES", default)

[StefanKarpinski]

That's what I had at first, but there are two issues:

1. I don't want to even try to make a symlink if the variable is set.
2. If the variable is set but has an invalid value, I don't want this code to inscrutably error.

The former would be addressed if the API to Base.get_bool_env allowed a callback to get a default value that is only evaluated if needed (like get). The latter would be addressed if Base.get_bool_env raised an error on invalid value instead of returning nothing.

[IanButterworth]

Sounds reasonable JuliaLang/julia#52950

[StefanKarpinski]

If someone wants to help me debug this, I'm slammed and haven't and won't have the time (kid was sick last week, I was sick over the weekend, childcare is sick now, plus have to work), but this needs to go into 1.10.1. An intermediate version did work, now it's failing tests, probably because the tests expect some specific error message. If someone could help track down why this is failing now, that would really help.

[IanButterworth]

Ok. The logic was just inverted on ignore_hash_env_set. This passes artifacts tests locally now

[StefanKarpinski]

Thank you so much, @IanButterworth! You're the best ❤️