Regression: LocalFileSystem now tries to resolve links

[isidentical]

Test Script

import contextlib
import os

import fsspec

def cleanup(*files):
    for file in files:
        with contextlib.suppress(FileNotFoundError):
            os.unlink(file)

cleanup("/tmp/test", "/tmp/link")


# Create a file, symlink it to another path and
# delete the original
with open("/tmp/test", "w") as stream:
    stream.write("hey")
os.symlink("/tmp/test", "/tmp/link")
os.unlink("/tmp/test")

fs = fsspec.filesystem("file")

print("(fsspec) /tmp/test exists:", fs.exists("/tmp/test"))
print("(fsspec) /tmp/link exists:", fs.exists("/tmp/link"))

print("(os) /tmp/test exists:", os.path.exists("/tmp/test"))
print("(os) /tmp/link exists:", os.path.exists("/tmp/link"))

With the script above, fsspec==2021.07.0 returns True for fs.exists() on a symlink where the original file it is linked to is deleted. After 63a668f, it now returns False since it tries to resolve the links via follow_links=True.

I personally find the change to make sense, considering this is how the os.path.exists behaves but I believe it should be explicitly mentioned in the changelog and as well as an option to opt-out from it. Perhaps a keyword option called follow_symlinks=... to the info and exists or a method called lexists() (similiar to os.path.lexists).

[martindurant]

Would be happy to see lexists added. I suppose it would only make any sense for local and maybe ssh.

[martindurant]

Is that an OK plan? Should we close this?

[isidentical]

The issue is not resolved yet, so I don't think we should close it. I'll try to come up with a patch