Re: BIND and write locks

Peter wrote on 09/01/2003 06:55:15 AM:

> I am trying to figure out, how write locks should behave WRT binding. 
> Take the following scenario: 
> - a collection C11 
> - a collection C1 containing a binding "foo" to C11 and mapped to URI 
"u1" 
> - a resource R2 
> - a collection C2 containing a binding "b" to R2 and mapped to URI "u2" 
>    u1| 
>      C1 
>   foo|    |u2 
>     C11   C2 
>           |b 
>           R2 
> Now the following requests are issued (' marks locked resources): 
> - LOCK /u1 (write, exclusive, depth=infinity) 
>   --> C11 is automatically added to the write-lock 
> - PUT /u1/foo/a (passing locktoken) 
>   --> creates R1 which is added to the write-lock (RFC2518, Section 7.5) 

> - BIND /u1/foo (b->/u2/b, passing locktoken) 
>   --> R2 is added to the write-lock (RFC2518, Section 7.5) 
>    u1| 
>      C1' 
>   foo|    |u2 
>     C11'  C2 
>     a| b\ |b 
>      R1'  R2' 
> - REBIND /u2 (a->/u1/foo/a, passing locktoken) 
>   --> R1 is removed from the write-lock (RFC2518, Section 7.7) 
>    u1| 
>      C1' 
>   foo|    |u2 
>     C11'  C2 
>        b\ |b \a 
>           R2' R1 
> The described behavior for BIND and REBIND is what I suppose it 
> should be. Is it correct? 

Yes.

> BTW, what happens if, afterwards, the following UNLOCK on R2 is issued: 
> - UNLOCK /u1/foo/b (passing locktoken) 
>   --> a) request is rejected, UNLOCK must be issued of C1 (/u1) 
>   or 
>   --> b) all associated resources (C1, C11, R2) are unlocked 
> (RFC2518, Section 8.10.4) 

Several of us have strongly advocated (a), to avoid a client mistakenly
unlocking a whole tree of resources when they intended to unlock a
single resource.  I don't believe this has been resolved though.

> I suppose, that it is not possible to remove single resources from 
> the write-lock by means of UNLOCK, isn't it? 

That is correct, it is not possible.

> P.S.: 
> At http://www.webdav.org/specs/ I found a link to http://ftp.ics.
> uci.edu/pub/ietf/webdav/collection/bind-issues.html, which contains 
> the following entry:
>    "ID: 41 
>    Source: Reuter/Hunt 
>    Description: Specify how BIND interacts with a write lock. 
>    Status: Closed 
>    Resolution: Declined 
>    Locking semantics is in too confused a state currently to 
>    be able to make any reliable statements. Don't want to hold 
>    up binding spec till lock settles down." 
> Is that still prevailing? 

No, that is an obsolete document, and it (and any links to it) should
be removed.  The current bind issues document is:

http://www.webdav.org/bind/bind-issues-list.htm

Cheers,
Geoff