Description
@ostronom @mziccard It seems #1277 has caused a regression in my code. I use / characters in the blob names, and they are now being unexpectedly encoded as %2F in the signed path.
It seems that object naming requirements are not equivalent to URL encoding.
When I round-trip a signed url, Storage#get now fails to find the successfully uploaded file (confirmed via gsutil and Cloud Console).
My understanding is that /s are valid characters for blob names. From the same page quoted above:
The canonical resource is everything that follows the host name. For example, if the Cloud Storage URL is https://storage.googleapis.com/example-bucket/cat-pics/tabby.jpeg, then the canonical resource is /example-bucket/cat-pics/tabby.jpeg.
It seems like this library needs to use a different escape function for object names. (Also, I noticed that the new test testSignUrlForBlobWithSpecialChars, does not cover slashes.)
Does this make sense? What is the intended behavior for blob names containing slashes?
repro test:
BlobInfo blobInfo1 = BlobInfo.builder(bucketName, "foo/bar/baz #%20other cool stuff.txt").build();
storage.create(blobInfo1, "content".getBytes());
URL url = storage.signUrl(
blobInfo1,
1,
TimeUnit.HOURS,
httpMethod(HttpMethod.PUT),
httpMethod(HttpMethod.GET)
);
final String[] split = url.getPath().split("/");
Blob notFound = storage.get(split[1], split[2]); // unexpectedly null!
Blob found = storage.get(split[1], URLDecoder.decode(split[2])); // found, surprisingly