Fix calls to realpath for shared network drives

[corakingdon]

see #33956 for discussion

[StefanKarpinski]

Looks good to me. Let us know when you've tested with the CI-built binary.

[fredrikekre]

Answering #33956 (comment) here:

Thanks @fredrikekre! where on the PR can I find the option to download?

https://s3.amazonaws.com/julialangnightlies/assert_pretesting/winnt/x64/1.4/julia-fe14bfcf42-win64.exe

[corakingdon]

Thank you! for future reference though, where did that link come from?

[KristofferC]

Click Details on the buildbot/tester_win64 (https://build.julialang.org/#/builders/32/builds/6845)

Then go to the Build properties tab and look at

download_url | "https://s3.amazonaws.com/julialangnightlies/assert_pretesting/winnt/x64/1.4/julia-fe14bfcf42-win64.exe"

[corakingdon]

@StefanKarpinski in combination with the changes made to Pkg, this seems to work fully on my network drive! I'm wondering though if this is a strange work around, and if it would be better to change the realpath definition instead, since it's a julia function? thoughts?

julia/base/path.jl

Lines 409 to 434 in 8707744

	"""
	realpath(path::AbstractString) -> String
	Canonicalize a path by expanding symbolic links and removing "." and ".." entries.
	On case-insensitive case-preserving filesystems (typically Mac and Windows), the
	filesystem's stored case for the path is returned.
	(This function throws an exception if `path` does not exist in the filesystem.)
	"""
	function realpath(path::AbstractString)
	req = Libc.malloc(_sizeof_uv_fs)
	try
	ret = ccall(:uv_fs_realpath, Cint,
	(Ptr{Cvoid}, Ptr{Cvoid}, Cstring, Ptr{Cvoid}),
	C_NULL, req, path, C_NULL)
	if ret < 0
	ccall(:uv_fs_req_cleanup, Cvoid, (Ptr{Cvoid},), req)
	uv_error("realpath", ret)
	end
	path = unsafe_string(ccall(:jl_uv_fs_t_ptr, Cstring, (Ptr{Cvoid},), req))
	ccall(:uv_fs_req_cleanup, Cvoid, (Ptr{Cvoid},), req)
	return path
	finally
	Libc.free(req)
	end
	end

[StefanKarpinski]

It's true that the realpath function is kind of problematic, but it's a very low level function and I don't think we want our realpath to deviate too much from the system one, which has this behavior. Honestly, in this case we maybe shouldn't be calling realpath at all. In any case, improving/changing realpath is a much bigger discussion and since this fixes your issue, let's go with it. We can merge the two fixes independently and then once they're both included in master, the issue will be fixed. Another question is whether these are backportable fixes or not. I guess I don't really see why not, but it would be good to get some other opinions on backporting.

[StefanKarpinski]

I've approved but, @ckingdon95, I need you to make this no longer a draft before I can merge.

[StefanKarpinski]

I've opened #34135 to discuss improving realpath or offering a better replacement for it.