Tilde in file name expanded incorrectly on Linux - works on Mac about artifactory HOT 8 CLOSED

BTW you're probably getting different behavior on macOS versus Linux because the two have different versions of urllib3. urllib3's URL parser is used with the context of an HTTP client so it optimizes for sending over the wire, not parsing URLs and passing them to application contexts (ie doesn't do the hex decoding)

from artifactory.

% in file name breaks both Mac and Linux. These characters are important for Debian packages such as libgcc1_1%3a8.3.0-6ubuntu1~18.04.1_amd64.deb

from artifactory.

relative_to function may be related https://github.com/devopshq/artifactory/blob/develop/artifactory.py#L759

from artifactory.

Found the problem with % on drv = part.split(url.path)[0]

url.path comes from urllib3.util.parse_url() which calls urllib3.util._encode_invalid_chars() which has IMO a bug that changes all percent-encoded bytes to upper case

# Normalize existing percent-encoded bytes.
for enc in percent_encodings:
    if not enc.isupper():
        component = component.replace(enc, enc.upper())

This modifies the URL's casing, and part-shuffling fails.
Opened an issue urllib3.util._encode_invalid_chars() changes the percent-encoded bytes to upper case #1677

from artifactory.

Same rc is also causing URLs with ~ to fail. During drv = part.split(url.path)[0], url.path is url-encoded while part is not.

387  ->	                drv = part.split(url.path)[0]
388  	                path_parts = url.path.strip(sep).split(sep)
389  	                root = sep + path_parts[0] + sep
390  	                rest = sep.join(path_parts[1:])
391  	                return drv, root, rest
(Pdb) url.path
'/something/something/containerd.io_1.2.0%7Erc.0-1_amd64.deb'
(Pdb) part
'https://domain.com/something/something/containerd.io_1.2.0~rc.0-1_amd64.deb'

In order for drv = part.split(url.path)[0] to work, either part must be encoded or url.path must be decoded.

Now, there seems to be a fight online about whether tilde should be encoded or not. I guess Linux version of urllib3 encodes it whereas Mac version of urllib3 does not? Since the dependent packages are yielding inconsistent results, I think ArtifactoryPath should program defensively and expect either cases.

from artifactory.

Welp that was fast. The issue i opened was closed within 8 hours as a "feature not a bug". Regardless, I think ArtifactoryPath should keep the original casing of percent-encoded bytes because otherwise the copy flat out fails mysteriously. A lowercase hex after an @ is a valid file name after all.

from artifactory.

Not sure what the best way to unquote the URL is. I'm writing something like

def _get_path(part, path):
    if path in part:
        return path

    unquoted_path = urllib3.parse.unquote(path)
    if unquoted_path in part:
        return unquoted_path
    else:
        raise ValueError(f"{part} does not include {path}")

but it seems that urllib3 does not have parse.unquote() which its counterpart urllib does. It seems silly to import urllib just to call unquote. Thinking...

from artifactory.

ubuntu 16.04 - All work correctly

>>> from artifactory import ArtifactoryPath
>>> target = ArtifactoryPath("https://artfactory/repo/tmp", auth=('user', 'password'))
>>> target.deploy_file('file~with~tilde')
>>> origin = ArtifactoryPath("https://artfactory/repo/tmp/file~with~tilde", auth=('user', 'password'))
>>> target = ArtifactoryPath("https://artfactory/repo/tmp/file~with~tilde2", auth=('user', 'password'))
>>> origin.copy(target)
>>> target_two = ArtifactoryPath("https://artfactory/repo/tmp/file~with~tilde~two", auth=('user', 'password'))
>>> origin.copy(target_two)

from artifactory.