W3C home > Mailing lists > Public > w3c-dist-auth@w3.org > January to March 1997


From: Yaron Goland <yarong@microsoft.com>
Date: Wed, 26 Mar 1997 00:10:38 -0800
Message-ID: <11352BDEEB92CF119F3F00805F14F485026B72C7@RED-44-MSG.dns.microsoft.com>
To: DAV Discussion <davdisc@microsoft.com>

1.1	Problem Description

Like checkout, check-in is a roundly abused method. In the same survey
on versioning systems mentioned previously, the DAV authors determined
that check-in is used for:
*	Specifying the resource from which the checked in resource is
*	Specifying a working location from which the server should
obtain the new version
*	Specifying the version identifier that should be associated with
the newly created resource
*	Linking in the newly created version to the version tree and
updating history

I believe that the features that check-in should provide are the ability
to specify a resource to be checked in, either by including it in the
check-in method or by referring to it in a header, and the ability to
specify the parent(s) and version identifier of the new resource.

1.2	Proposal

In the same vein as my previous proposal, I propose to be rid of the
CHECKIN method. My observation is that the only relationship between
check-in and checkout, is the implied permission that a checkout gives
to perform a check-in. This permission is maintained through the use of
the lock method, even if the lock is advisory.
Given that the primary duty of Check-In is to create a new file I
propose providing the creation facility through PUT.
In order to achieve check-in's full functionality it is necessary to use
the Merge, Source, and Version-ID headers.
The Source header, which I have previously defined, is used to create a
new resource by reference rather than by value. This functionality is
identical to copy, hence the use of the Source header.
The Merge header, defined separately, is used to indicate that the
resource to be created is the result of a merger between the specified
URIs and, if included, the request body.
The Version-ID header, defined in this section, is used to request or
define the version identifier to be associated with this resource. 

1.3	Version-ID Header

VersionID = "Version-ID" ":" ReqType VID
ReqType = "Request" | "Demand" | "Set" | Token
VID = "<" Token ">" | Quoted-String

The Version-ID header may be used to either request or demand a version
identifier. If the "Request" attribute is used then the server may
freely ignore the version-id header. However, if the "Demand" header is
used then the server may only accept the PUT if it will assign the
specified version identifier to the new resource. If the server can not
comply then it should return a 412 Precondition Failed.

If the server wishes to inform the client of the version identifier
assigned to their resource it can return the Version-ID header with the
Set attribute.

Note: The reason I have chosen a new header rather than using
Content-Version is because of the need to specify the strength of the
request. However it is equally acceptable to use the Content-Version
header along with a second header to specify the strength of the
Content-Version request.

1.4	Where to Check In To?

Most versioning systems do not support arbitrary check-in locations when
creating a new version of an existing resource. To handle this situation
I propose that the default behavior when creating a new version of a
resource is to perform the PUT on the tree handle of the resource with
the merge headers set to the parents. If successful a 201 "Created"
response will be returned with a Location header containing the URI of
the new version.

In the case of a truly distributed system the client may create the new
version wherever they desire but there will be no guarantee that the
various versioning headers will be properly maintained. The client will
have to check the relevant resources and if the headers have not been
properly set, the client will have to use the LINK method to set and
maintain them itself.

1.5	Requiring Check Out

Many systems will only allow a new version of a resource to be created
if its parent is checked out. Rather than introducing a new result that
says "You must lock this resource in order to create a new version" I
propose we make a LOCK/UNLOCK of the appropriate type implicit in the
creation of a new version of a resource. The LOCK will only last for the
duration of the PUT. Furthermore, if the client is PUTting to a tree
handle it MUST be prepared to accept a LOCK error response in case there
is a problem.
Received on Wednesday, 26 March 1997 03:22:59 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 20:01:10 UTC