Hi Spencer,
I've updated my copy of the draft, so I am answering the remaining
comments you've sent:
Spencer Dawkins wrote:
Abstract
Many IMAP clients use the result of a SEARCH command as the input to
perform another operation, for example fetching the found messages,
deleting them or copying them to another mailbox.
This can be achieved using standard IMAP operations described in RFC
3501, however this would be suboptimal: the server will send the list
of found messages to the client, after that the client will have to
parse the list, reformat it and send it back to the server. The
client can't pipeline the SEARCH command with the subsequent command.
Spencer (readability) - in this paragraph, there are three ways that
this extension improves normal operation - for bandwidth (because the
list isn't sent in both directions when it doesn't have to be), for
processing (since the list isn't parsed and reformatted when it
doesn't have to be), and for serialization (since the two commands can
now be pipelined). In the body of the document, a fourth improvement
is also mentioned ("server optimization"), but isn't described at all.
The two lists should be the same, and I wouldn't mind seeing some
description of "server optimization" in the body of the document.
Done.
This document proposes an IMAP extension that allows a client to tell
a server to use the result of a SEARCH (or UID SEARCH) command as an
input to any subsequent command.
2. Introduction and Overview
Spencer (readability) - This section combines introduction/overview (not
normative) and protocol operation (normative). I'd suggest separating the
two. I'd also suggest using some of the abstract text as an actual
introduction - this section assumes that you've read the abstract, I
think,
and not everyone will automatically do that.
Done.
The SEARCH result reference extension defines a new SEARCH result
option [IMAPABNF] "SAVE" that tells the server to remember the result
of the SEARCH or UID SEARCH command (as well as any command based on
SEARCH, e.g. SORT and THREAD [SORT]) and store it in an internal
variable that we will reference as the "search result variable". The
client can use the "$" marker to reference the content of this
internal variable. The "$" marker can be used instead of message (or
UID) sequence in order to indicate that the server should substitute
Spencer (nit) - "message" looks like a noun until you get past "(or UID)"
and discover it's an adjective. Suggest s/message (or UID)
sequence/message
sequence or UID sequence/.
Changed.
it with the list of messages from the search result variable. Thus
the client can use the result of the latest remembered SEARCH command
as a parameter to another command. The search result marker has
several advantages:
* it avoids wasted bandwidth and associated delay;
* it allows the client to pipeline a SEARCH [IMAP4] command with a
subsequent FETCH/STORE/COPY/SEARCH [IMAP4] or UID EXPUNGE
[UIDPLUS] command;
* the client doesn't need to spend time reformatting the
result of a SEARCH command into a message set used in
the subsequent command;
* it allows the server to perform optimizations.
Spencer (readability) - in this list, there are FOUR advantages, not
three as above... can you give any information about what
optimizations a server can perform because this extension is being used?
I've added the following text:
For example if the server
can execute several pipelined commands in parallel (or out of order),
presence of the search result marker can allow the server to decide
which commands may or may not be executed out of order.
In absence of any other SEARCH result option, the SAVE result option
also suppresses any SEARCH response that would have been otherwise
returned by the SEARCH command.
Spencer (readability) - somewhere around here, the actual protocol
specification section starts... :-)
I've split the section.
[...]
command with the SAVE result option that caused the server to return
NO tagged response sets the value of the search result variable to
the empty sequence.
When a message listed in the search result variable is EXPUNGEd, it
is automatically removed from the list. Implementors are reminded
that if the server stores the list as a list of message numbers, it
MUST automatically adjust them when notifying the client about
Spencer (readability) - is the meaning of "automatically adjust them"
obvious to everyone but me? Do you just remove the message number from
the
list, or renumber the list, or ... ?
The latter. This is described in RFC 3501, section 7.4.1. I've added
this reference to clarify.
expunged messages.
Note that even if the "$" marker contains the empty list of messages,
it must be treated by all commands accepting message sets as
parameters, as a valid, but non matching list of messages. For
Spencer (readability) - this is correct as written, but hard to parse.
Perhaps "treated as a valid but non matching list of messages, by all
commands that accept message sets as parameters"?
I like that. Thanks.
example, the "FETCH $" command would return tagged OK response and no
FETCH responses. See also the Example # 5 below.
[...]
2) The following example demonstrates that the result of one SEARCH
command can be used to subset the result of another SEARCH
command:
Example 3:
C: A300 SEARCH RETURN (SAVE) SINCE 1-Jan-2004
NOT FROM "Smith"
S: A300 OK SEARCH completed
C: A301 UID SEARCH UID $ SMALLER 4096
S: * SEARCH 17 900 901
S: A301 OK completed
Note that the second command in Example 3 can be replaced with:
C: A301 UID SEARCH $ SMALLER 4096
and the result of the command would be the same.
Spencer (readability) - I didn't quite follow this. I think what
confused me
is the introductory text - isn't this example demonstrating "that the
result
of one SEARCH command can be used as input to a second SEARCH command"?
"Subset" as written seemed to say that the first result was taking an
action
("subset"), but it's just input to the second SEARCH...
I've changed the text as you've suggested.
2.2. Multiple Commands in Progress
Spencer (readability) - could you insert a sentence that introduces
Example
7 and explains what you illustrate in this example? (Is it only what the
//comment says?) (same for Example 8)
Example 7:
C: G282 SEARCH RETURN (SAVE) KEYWORD $Junk
C: G283 SEARCH RETURN (ALL) SINCE 28-Oct-2006
FROM "Eric"
//The server can execute the two SEARCH commands
//in any order, as the don't have any dependency.
Spencer (nit) - s/the/they/
Fixed.
//Note that the second command is making use of
//the [ESEARCH] extension.
S: * ESEARCH (TAG "G283") ALL 3:15,27,29:103
S: G283 OK SEARCH completed
S: G282 OK SEARCH completed
Example 8:
C: H282 SEARCH RETURN (SAVE) KEYWORD $Junk
C: H283 SEARCH RETURN (SAVE) SINCE 28-Oct-2006
FROM "Eric"
//The result of the second SEARCH overrides the
//result of the first.
S: H283 OK SEARCH completed
S: H282 OK SEARCH completed
2.3. Interaction with ESEARCH extension
When the SAVE result option is combined with the MIN or MAX
[ESEARCH] result option, and none of the other ESEARCH result
options are present, the corresponding MIN/MAX is returned (if
the search result is not empty) but the "$" marker would contain
a single message as returned in the MIN/MAX return item. If the
SAVE result option is combined with both MIN and MAX result
options, and none of the other ESEARCH result options are
present, the "$" marker would contain one or two messages as
returned in the MIN/MAX return items. If the SAVE result option
is combined with ALL and/or COUNT result option, the "$" marker
would always contain all messages found by the SEARCH or UID
SEARCH command. (Note that the last rule might affect ESEARCH
implementations that optimize how COUNT result is constructed.)
Spencer (readability) - is it obvious to everyone except me *how* the
last
rule might affect these implementations?
Spencer (nit) - the previous paragraph is pretty dense. It might be more
readable as a bulleted list (one bullet per case).
I've split the paragraph into multiples. I hope this is more readable.
3. Formal Syntax
4. Security Considerations
This extension requires the server to keep additional state, that may
be used to simplify Deny of Service attacks. In order to minimize
damage from such attacks server implementations MAY limit the number
of saved searches they allow across all connections at any given time
and return the tagged NO response to a SEARCH RETURN (SAVE) command
when this limit is exceeded.
Spencer (readability) - I'm guessing there is no way for a client to
discover this is the reason for the tagged NO response?
I've added a new IMAP response code to allow the server to signal that
its limit is exceeded.
but it might be nice
to say so explicitly. Just curious - is this what an IMAP server would do
today, when it detects a DoS attack (by an attacker who's just working
harder)? If so, it might be nice to point that out, too.
5. IANA Considerations
This document defines the "X-DRAFT-I05-SEARCHRES" <<Fix upon
publication>> IMAP capability. IANA is requested to add it to the
IMAP4 Capabilities Registry, which is currently located at:
http://www.iana.org/assignments/imap4-capabilities
Spencer (nit) - IANA has suggested "search-res" as the permanent name
for this capability. You might add it here, just for completeness.
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf