How to copy a specified version of an object?

[christiangenco]

The Amazon REST API Documentation cites the following:

Sample Request: Copying a specified version of an object

The following request copies the key my-image.jpg with the specified version ID and copies it into the bucket bucket and gives it the key my-second-image.jpg.

PUT /my-second-image.jpg HTTP/1.1
Host: bucket.s3.amazonaws.com
Date: Wed, 28 Oct 2009 22:32:00 GMT
x-amz-copy-source: /bucket/my-image.jpg?versionId=3/L4kqtJlcpXroDTDmJ+rmSpXd3dIbrHY+MTRCxf3vjVBH40Nr8X8gdRQBpUMLUo
Authorization: authorization string

I see the copy_from and copy_to methods in Aws::S3::Object, but I can't for the life of me figure out how to specify a versionId in either of these requests.

Am I missing something painfully obvious, or should I resort to a raw REST request?

[christiangenco]

Here's one of the more probable things I've tried:

source_object = Aws::S3::ObjectVersion.new({
  bucket_name: "source_bucket",
  object_key: "test.txt",
  id: "WYIIJB3UpKixITFbqLP6nVbAJsa5Ul5F",
  client: Aws::S3::Bucket.new(name: "source_bucket").client # there's probably an easier way to get a client object
})

destination_object = Aws::S3::Bucket.new("destination_bucket").object("test.txt")

source_object.copy_to(destination_object) 
# NoMethodError: undefined method `copy_to' for #<Aws::S3::ObjectVersion:0x007fb7f9052d88>

destination_object.copy_from(source_object) 
# ArgumentError: expected source to be an Aws::S3::Object, Hash, or String
# from .rbenv/versions/2.1.1/lib/ruby/gems/2.1.0/gems/aws-sdk-resources-2.1.30/lib/aws-sdk-resources/services/s3/object_copier.rb:44:in `copy_source'

[christiangenco]

Looks like that copy_from didn't work because S3::ObjectCopier#copy_source doesn't support S3::ObjectVersion as a possible argument (and unlike version 1, S3::ObjectVersion is not a subclass of S3::Object).

I thought the #copy_source method might be monkey patched to fix this:

module Aws
  module S3
    class ObjectCopier
      def copy_source(source)
        case source
        when String then escape(source)
        when Hash then "#{source[:bucket]}/#{escape(source[:key])}"
        when S3::Object then "#{source.bucket_name}/#{escape(source.key)}"
        when S3::ObjectVersion then "#{source.bucket_name}/#{escape(source.key)}?versionId=#{escape(source.id)}"
        else
          msg = "expected source to be an Aws::S3::Object, AWS::S3::ObjectVersion, Hash, or String"
          raise ArgumentError, msg
        end
      end
    end
  end
end

but now destination_object.copy_from(source_object) throws:

NotImplementedError: #load not defined for Aws::S3::ObjectVersion
from .rbenv/versions/2.1.1/lib/ruby/gems/2.1.0/gems/aws-sdk-resources-2.1.30/lib/aws-sdk-resources/resource.rb:150:in `load'

I can't find an actual definition for Aws::S3::ObjectVersion (it looks like it's being dynamically created from Aws::Resources::Resource?), so I'm not sure how to implement this mysterious #load method. It might just be assigning a LoadOperation, but I'm in over my head at this point. The only way I see forward at the moment is to define Aws::S3::ObjectVersion identically to how Aws::S3::Object is currently defined and hope the load operation sorts itself out.

I'll start down the road of submitting a raw REST request, though figuring out how to sign requests outside of this gem looks like it might be just as difficult.

[trevorrowe]

This is currently a limitation of the #copy_from and #copy_to methods. As a workaround you use the Aws::S3::Client#copy_from method directly:

s3 = Aws::S3::Client.new
s3.copy_from({
  bucket: 'target-bucket',
  key: 'target-key',
  copy_source: "source-bucket/source-key?versionId=123456789..."
})

You need to ensure the source key is correctly escaped. You can call Seahorse::Util.uri_path_escape(key) to ensure the key is correctly escaped. You don't have to worry about the bucket name, S3 limits bucket names to characters that do not require escaping. You might have to call Seahorse::Util.uri_escape on the version id value. I don't know what characters might appear in the version ID, but escaping the ID should be perfectly reasonable.

I'll mark this issue as a feature request to add support for copying versioned IDs to the resource interfaces as a feature request.

[christiangenco]

Ahh perfect! Thank you so much 😄

For anyone else walking down this path, check out the full documentation on that Aws::S3::Client#copy_object method.

Also, it looks like you just added Seahorse::Util#uri_path_escape and it hasn't been published to rubygems yet, so I just patched it in an init script:

if !defined? Seahorse::Util.uri_path_escape
  module Seahorse
    module Util
      class << self
        def uri_path_escape(path)
          path.gsub(/[^\/]+/) { |part| uri_escape(part) }
        end
      end
    end
  end
end

Those are some nifty helper methods - URI escaping is always strangely confusing in Ruby. It'd be lovely if you could just do String#escape or String#escape_url, but I digress.

[awood45]

Added this to the feature request backlog.

[lwoggardner]

Not sure how to "vote" for feature requests, so I'll just "me too" here.

[mullermp]

Reopening - deprecating usage of Feature Requests backlog markdown file.