I generally agree with Pete's comments too. I also think that the specication should provide import and export functions which might support multiple formats. The specification could mandate at least one be supported and others as optional. For instance, it might be good to be able to put/get the contents with compression activated. Just an idea. I would also put the concept of "patches" on the table as it might impact the specification. Keisuke Fukui wrote:
Continueing from the previous post...
For the rest of comments from Pete, Ziu, Peter wrote:
In response to your questions:
I'd like to ask everyone which of the below is the real issue we are facing: - Name for the things to be registered; i.e. AA file in my diagram, or Archive description and Application contents file(s) (includes DD) in Tom's one. If this is an issue, I can be so much flexible about how we can call it (them).
I am not particularly hung up on names right now, that is a refinement issue, and not a defining structure. The issue is whether there should be a porttype which has a method for pushing an entire and complete physical AA into a repository.
So do I.
Here is my take on the issue:
- In the case of an ACS implementation, sending an entire AA imposes a need to "know" about a physical file format. This is contradictory to our statement that "we will not define how an ACS will store it's data".
We will define the archive format, but will not break into the data representation to each of the file contents. I think it's like a defineing TCP packet or SOAP envelope without specifying the format or representation of the payload data.
For example, if my implementation of ACS requires a method in the porttype that requires the acceptance of a physical AA file, then I must parse the AA file into parts in order to store it into the format that my implementation has chosen. Perhaps my AA file is stored in thousands of DB2 or Oracle records, since using one record to store an AA file will most likely fail at some point when too big.
Using the off-the-shelf RDBMS is one of the posibilities and how the record is designed to accommodate the AA file content is out of scope of the specification, just like the internal representation of the repository. Do you believe we need to break into this?
If I had porttypes that allowed me to remotely construct an AA, we can punt on a physical format and an entire physical AA transmission format. Providing we have the appropriate porttypes to remotely construct, inspect, and retrieve parts of an AA, then we only need to define on the wire how to move parts of an AA around, yet we can still maintain the notion of an AA and it's integrity, since the interface that constructed it can enforce it's integrity. This is why Tom has outlined sending the entire AAD (hope that the correct acronym) when replicating. The receiver of the replica can inspect and traverse the AAD to find the parts to pull the file over in chunks... especially if there was "modified date" time stamp info on each individual part, then we could explore the replication model I outlined.
Yes we can define only the Descriptor in XML doc, and we can additionally specify the packet or envelope of the data; this will make easier for future IDEs comform to the ARI and AAF. If we are discussing we should only have ARI but not AAF, it may be feasible, but as soon as we specify the AAD schema, it is a part of the current AAF. So I guess the residue here is if we were to define the physical format in addition to the AAD schema.
- A physical format of the archive; a zip is described in the strawman. If this is an issue, we can focus on this topics. However, it is not mentioned as this so far.
- I think this is what Tom has been trying to focus on in his .ppt last meeting. I seemed to have put my physical file comments in the previous question, sorry.
Yes this is my take, too. The only reason why I didn't hit this was no one pointed this obvious argument so far, if it is.
- Partial update of the archive in the repository (It seems to me a separate issue that can be discussed independently.)
- it is related since defining how parts of an AA are created, accessed, modified, and removed have implications that, in the case of physical file formats, would need to be known by ACS if an ACS impl. decided to just write the received AA file to disk, and then later have to modify it.
Yes, it is related, but I thought still we can discuss this at another occasion.
- Assuming IDE products to be bound to ACS specification (IMHO, it's desirable, but the decision remains up to the vendors.)
I am not sure I understand this question. No matter what porttype spec we come up with, as long as we do not create any issues for the IDE's or ant (or any other client type), we should be ok. I think that having the client create one file will have scalability issues to remote clients of ACS for large AAs.
I neither have conceivable reason for this. However, I remember Tom appealed there is (or should) not such IDEs in the previous conversation.
- Or anything else?
That's my take.
I listed as many as topics I can imagine in my post, and can't imagine any more at the moment w.r.t. the AA creation use case.
I appreciate comments from anyone.
- Keisuke
-- Michael Behrens R2AD, LLC (571) 594-3008 (cell) *new* (703) 714-0442 (land)