Hi Michael and Tom,
My comments are below, preceded by 'ira>'.
Please see in particular at the end my comments on the
use of 'collection' for the new 'printer-xri-supported'
attribute that 'repairs' our previous mistake in IPP
with the three musketeers.
Cheers,
- Ira McDonald, consulting architect at Sharp Labs America
High North Inc
-----Original Message-----
From: Michael Sweet [mailto:mike at easysw.com]
Sent: Wednesday, March 22, 2000 5:55 AM
To: Hastings, Tom N
Cc: ipp
Subject: Re: IPP> FW: Thoughts on the new Move-Job operation
"Hastings, Tom N" wrote:
>> Michael, Ira, Bob, and I have been exchanging email on the Move-Job
> operation as a result of last week's IPP telecon. We have a few
> issues left. But here is where we are for tomorrow's IPP telecon,
> 3/22.
As usual, I won't be able to "attend" the telecon... :(
My comments are below...
> ...
> ISSUE 01: There is some debate as to whether to ALLOW the Move-Job
> operation to be supported when the job is in the 'processing' state.
> If it is allowed, it would be a MAY, not a MUST, because some
> systems will have problems with accounting if the same job-id is
> reused for the job again if some resources had been consumed.
See my other comments on this; to summarize, I think we'll need to
allow the "move-job" operation to create a new job-id and job-uri
as needed by the implementation. The new job-id should not be
REQUIRED, since this will open up another can of worms with
accounting and job persistence - e.g. doubling the server's disk/
memory requirements if document files are persistent until purged.
(something that CUPS 1.1 supports)
ira> I agree with the caveat that the reused job-id MUST represent
ira> a job which NEVER entered the 'processing' state on the original
ira> Printer - otherwise it becomes an avenue for an accounting
ira> exploit that runs a job twice and gets charged once.
> ...
> 2. In case the Printer defaults are different for the new Printer,
> we need to specify that the new Printer's defaults will be used when
> the job is processed, even if they differ from the defaults of the
> old Printer.
This makes sense, since in the absense of job template attributes the
printer defaults (which the client may be oblivious to) are used
anyways by Create-Job, Send-Job, and Send-URI.
ira> I agree.
> ...
> "printer-uri" (or old "job-uri") and the new "printer-uri".
Which should be called "job-printer-uri" to avoid ambiguitity with
the printer-uri used to identify the job.
ira> Not sure which Printer URI is being renamed above. I'd
ira> suggest that an operation attribute in 'Move-Job' be called
ira> 'target-printer-uri' or 'new-printer-uri' for clarity.
> ...
> ISSUE 02: Ok to REQUIRE that the "ipp-attribute-fidelity" operation
> attribute be copied to the Job object, if the Move-Job operation is
> supported?
Yes. Similarly, if the new printer object does not support the
attributes provided, and ipp-attribute-fidelity is true, then
a client-error-conflicting-attributes error needs to be returned
and the job is not moved.
ira> I agree.
> 6. Finally, do we want to make Move-Job be like the Job Creation
> operations and specify that the Move-Job response MUST be the same
> as the Print-Job response:
Yes.
ira> I agree.
> ...
> ISSUE 03: Ok that Per-Job Subscriptions are automatically updated to
> be for the new job (whether the job-id changes or not)?
This is a sticky problem; if the job-id (and job-uri) changes, then
the recipient of the notifications may not know what the notification
is for (e.g. I am subscribed to job 5, I move the job, now I am
subscribed to job 6???)
Obviously we'll need a "move-job" event subscription, and that
event needs to provide the new job-id, job-printer-uri, and
job-uri attributes for the job (whether the job-id has changed or
not)
ira> This is covered by the notification content including the
ira> 'subscriber-user-data' opaque element (intended for client
ira> use to specify a useful correlation handle). In the Job
ira> Monitoring MIB we have the (normally client constructed)
ira> 'jmJobSubmissionID' for reliable correlation. In IPP
ira> notifications we also have 'job-name' (client supplied)
ira> in the standard bindings, which could be used for client
ira> correlation of the 'old' and 'new' jobs and their events.
> ISSUE 04: Should there be a new 'job-moved' event or is moving a
> job, just another operation that generates the 'job-created' (along
> with Print-Job, Print-URI, and Create-Job)?
I think we need it. If we end up requiring a new job-id (something
I'd rather not do), then we also need to add a new job-state value
for "job-moved", since "completed", "cancelled", and "aborted" do
not make sense.
ira> I agree that we need 'job-moved' as an event AND also in
ira> 'job-state-reasons'. We MUST NOT add a new 'job-state'.
ira> This would break all existing IPP and Job Monitoring MIB
ira> implementations. The Xerox MFP I worked with in the past
ira> on this feature transitioned the 'old' job to 'job-state'
ira> of 'cancelled' and 'job-state-reasons' of 'job-moved'.
> 10. ISSUE 05: For all of us to consider:
>> Should we add this operation to the Set Job and Printer Spec
> (because it is similar to scope and usage to the Set-Job-Attributes
> and Set-Printer-Attribute spec), add it to the Administrative Set2
> spec, or keep it as a separate spec?
It might make sense to include it there. However, I think we've
identified enough issues that move-job may be large enough to make
it a separate spec all by itself.
ira> I think we should add 'Move-Job' to the existing IPP Admin
ira> Operations spec (aka 'set2' which was a terrible name...).
ira> It does NOT belong in the IPP Set Operations spec.
> ...
> spec to go out for last call soon (as soon as we agree on the
> collection syntax encoding). On the other hand, if we are far from
> ...
Once again, let me voice my *lack* of support for the collection
syntax, as it will needlessly complicate implementation and support
of IPP. The new "printer-xri-supported" attribute could use a new
compound value tag ("xri") that would provide a URI and two keywords
to cover what is needed for that attribute (and other similar
situations we're likely to run into)
ira> Thanks Michael - I personally think we've already got 'complex'
ira> syntaxes for such things as resolution and that 'xri' would
ira> make a dandy base syntax to add to IPP.
ira> I'm VERY unhappy with the use of 'collection' syntax to add
ira> 'printer-xri-supported' attribute to IPP/1.x.
--
______________________________________________________________________
Michael Sweet, Easy Software Products mike at easysw.com
Printing Software for UNIX http://www.easysw.com