@@ -822,6 +822,27 @@ class S3(object):

     def object_copy(self, src_uri, dst_uri, extra_headers=None,

                     src_size=None, extra_label="", replace_meta=False):

+        """Remote copy an object and eventually set metadata

+

+        Note: A little memo description of the nightmare for performance here:

+        ** FOR AWS, 2 cases:

+        - COPY will copy the metadata of the source to dest, but you can't

+        modify them. Any additional header will be ignored anyway.

+        - REPLACE will set the additional metadata headers that are provided

+        but will not copy any of the source headers.

+        So, to add to existing meta during copy, you have to do an object_info

+        to get original source headers, then modify, then use REPLACE for the

+        copy operation.

+

+        ** For Minio and maybe other implementations:

+        - if additional headers are sent, they will be set to the destination

+        on top of source original meta in all cases COPY and REPLACE.

+        It is a nice behavior except that it is different of the aws one.

+

+        As it was still too easy, there is another catch:

+        In all cases, for multipart copies, metadata data are never copied

+        from the source.

+        """

         if src_uri.type != "s3":

             raise ValueError("Expected URI type 's3', got '%s'" % src_uri.type)

         if dst_uri.type != "s3":

@@ -837,8 +858,12 @@ class S3(object):

                 acl = None

         multipart = False

-

         headers = None

+

+        if extra_headers or self.config.mime_type:

+            # Force replace, that will force getting meta with object_info()

+            replace_meta = True

+

         if replace_meta:

             src_info = self.object_info(src_uri)

             headers = src_info['headers']

@@ -865,9 +890,8 @@ class S3(object):

                 threshold = self.config.multipart_copy_chunk_size_mb * SIZE_1MB

             if src_size > threshold:

-                # Sadly, s3 is badly done as metadata will not be copied in

-                # multipart copy unlike what is done in the case of direct

-                # copy.

+                # Sadly, s3 has a bad logic as metadata will not be copied for

+                # multipart copy unlike what is done for direct copies.

                 # TODO: Optimize by re-using the object_info request done

                 # earlier earlier at fetch remote stage, and preserve headers.

                 if src_headers is None:

@@ -883,6 +907,7 @@ class S3(object):

         else:

             headers = SortedDict(ignore_case=True)

+        # Following meta data are updated even in COPY by aws

         if self.config.acl_public:

             headers["x-amz-acl"] = "public-read"

@@ -898,6 +923,7 @@ class S3(object):

             headers['x-amz-server-side-encryption-aws-kms-key-id'] = \

                 self.config.kms_key

+        # Following meta data are not updated in simple COPY by aws.

         if extra_headers:

             headers.update(extra_headers)