Copy to non-existing S3 folder path creates a file with "/" instead of file name in that folder

[quantori-pokidovea]

Steps to reproduce:

# take an existing file
src = S3Path("s3://my-bucket/file.txt")
# and copy it into non-existent folder
dst = S3Path("s3://my-bucket/destination/")

src.copy(dst)

Current behavior

Creates a file with name / inside the folder s3://my-bucket/destination/

Expected behavior

The file s3://my-bucket/destination/file.txt is created

[pjbull]

FWIW .copy is meant to mimic the behavior of the copy function from shutil, which has similarly potentially unintuitive behavior (it both strips the delimiter and writes dst as a file).

from pathlib import Path
from shutil import copy

src = Path("text_file.txt")
src.write_text("hello")
#> 5

dst = Path("new_target/")

copy(src, dst)
#> PosixPath('new_target')

print(dst.is_file())
#> True
print(dst.read_text())
#> hello

Since this is CloudPath only API that is not on a Path object, we do have some leeway on deciding if there is a better approach. However, there is some ambiguity in the right approach since s3://bucket/file-end-in-slash/ is a valid name for a blob/object on at least some of the providers.

Some options:

• (as suggested above) Treat CloudPaths ending in / as directories in copy even if they don't exist
• Add a kwarg/setting to treat CloudPaths ending in / as directories in copy explicitly
• Add a warning if we ever write to a CloudPath ending in / (since this is likely user error)
• Leave as-is and document

If we document, the workaround is to concat the paths yourself (since not all providers even support empty folders):

src = S3Path("s3://my-bucket/file.txt")
dst = S3Path("s3://my-bucket/destination/")

src.copy(dst / src.name)