Re: [calsify] Benjamin Kaduk's Discuss on draft-ietf-calext-ical-relations-09: (with DISCUSS and COMMENT)

[Francesca Palombini <[e-mail]>]

Indeed, thank you very much Ben for the thoughtful review.

Mike: I will mark this as “revised ID needed” while waiting for the last (final) update with this minor change, then when v-11 is posted I’ll send this forward to the RFC Editor.

Thanks for the good work!
Francesca

From: iesg <[e-mail]> on behalf of Michael Douglass <[e-mail]>
Date: Monday, 7 March 2022 at 21:40
To: Benjamin Kaduk <[e-mail]>, Michael Douglass <[e-mail]>
Cc: [e-mail] <[e-mail]>, The IESG <[e-mail]>, [e-mail] <[e-mail]>, [e-mail] <[e-mail]>
Subject: Re: [calsify] Benjamin Kaduk's Discuss on draft-ietf-calext-ical-relations-09: (with DISCUSS and COMMENT)
Thanks Ben.

Your comments were very helpful.

I'll probably make at least one change along the lines suggested below
but I'll wait till after the calext session

On 3/7/22 14:38, Benjamin Kaduk wrote:
> Hi Michael,
>
> Most of this looks really good, and I'll go lift my Discuss shortly.
> Just a handful of notes scattered inline...
>
> On Sun, Feb 27, 2022 at 06:10:37PM -0500, Michael Douglass wrote:
>> Thank you for your detailed response - please see below ...
>>
>> On 2/15/22 17:26, Benjamin Kaduk via Datatracker wrote:
>>
>>> ...
>>>
>>>
>>> ----------------------------------------------------------------------
>>> DISCUSS:
>>> ----------------------------------------------------------------------
>>>
>>> The ABNF for linkparam (§8.2) incorporates a "langparam" production, but
>>> that is not defined in any of this document, RFC 5455, RFC 8288, or RFC
>>> 7986.  We need to define it somehow, whether by reference or directly.
>>> RFC 5545 does define a LANGUAGE parameter (our prose references a "LANG"
>>> parameter) and languageparam ABNF production, which is perhaps the
>>> simplest explanation for what was intended.
>> Yes - it should have been languageparam - corrected
> Excellent, I'm glad it was the easy fix.
>
>>> ----------------------------------------------------------------------
>>> COMMENT:
>>> ----------------------------------------------------------------------
>>>
>>> A couple broad comments before getting into the section-by-section
>>> details:
>>>
>>> I'm a little unclear on the value gained by having RELTYPE=REFID and
>>> RELTYPE=CONCEPT.  If the semantics were just supposed to be "is a member
>>> of the indicated group", then why do we need a relation type to say so.
>>> But if there's some other semantics associated, where do we define those
>>> semantics?
>> The idea is to associate one entity with others without that entity
>> necessarily having the associated attribute. The semantics are more like
>> - this is an aggregator for those things with the designated REFID or
>> CONCEPT
> Ah, thanks for the extra explanation.  Perhaps I was just being a bit dense
> when I first read it; I don't really have any suggestions for how we might
> add more text that would have prevented my confusion.
>
>>> There are a number of places (e.g., §1.4) where this document uses the
>>> term "collection" in a context that suggests that it should be a defined
>>> term of iCalendar, corresponding roughly to a "site" or administrative
>>> domain, but I didn't find any usage in RFC 5545 that would correspond to
>>> what's being described here.  Can we say more about what level of grouping
>>> this is intended to refer to?
>> More of a CalDAV thing. I'll precede the first occurrence with some text
>>
>> ------ OLD -------
>>
>> It is also often necessary to reference calendar components in other
>>
>> collections.  For example, a VEVENT might refer to a VTODO from which
>>
>> ------ NEW -------
>>
>> Calendar components are often grouped into collections to represent a
>> calendar or a series of tasks, for example [RFC4791] (CalDAV) calendar
>> collections.
>>
>> It is often necessary for calendar components in one such collection
>> to reference components in other collections.  For example, a VEVENT
>> might refer to a VTODO from which...
>>
>> ------------------
>>
>>> Section 1.1
>>>
>>>      The iCalendar [RFC5545] RELATED-TO property has no support for
>>>      temporal relationships as used by standard project management tools.
>>>
>>> I am admittedly not really the target audience of this specification, but
>>> I have no idea what the "standard project management tools" are that are
>>> being referred to.  Would (informative) references to a handful be
>>> appropriate?
>> I guess I should drop 'standard' and just write
>>
>>      temporal relationships as used by project management tools.
>>
>>
>> It's really the relationships which are 'standard' in the sense they are
>> apparently used by all/most project management tools.There is also the
>> Project Management Institute which has /A Guide to the Project
>> Management Body of Knowledge /which apparently defines these terms also.
>> I'm not sure if we want a reference to that - I think it's as close as
>> we get to a standard
>> //
>>
>>> Section 1.2
>>>
>>>      REFID is used to identify a key allowing the association of
>>>      components that are related to the same object and retrieval of a
>>>      component based on this key.  [...]
>>>
>>> This says "related to the same *object*" (emphasis mine), but the rest of
>>> the section just talks about an unstructured grouping.  It seems like we
>>> could give some hint of the nature of the "object" to which all the
>>> elements of the group are related, prior to the RELTYPE=REFID definition
>>> in §5.
>> Not sure how to reword this. The examples were an attempt to define
>> that. I should probably at least stick with component instead of object.
>>
>> I've tried a bit of rewording...
>>
>> ------ OLD -------
>>
>> REFID is used to identify a key allowing the association of
>> components that are related to the same object and retrieval of a
>> component based on this key.  Two examples of how this may be used
>> are to identify the tasks associated with a given project without
>> having to communicate the task structure of the project, and to group
>> all tasks associated to a specific package in a package delivery
>> system.
>>
>> ------ NEW -------
>>
>> REFID is used to identify a key allowing the association of
>> components that are all related to the referring, aggregating component and
>> the retrieval of components based on this key.  For example, this may be used
>> to identify the tasks associated with a given project without
>> having to communicate the task structure of the project. A further example
>> is the grouping of all sub-tasks associated with the delivery of a specific
>> package in a package delivery system.
>> ------------------
>>
>>> Section 1.3
>>>
>>>      The current [RFC5545] CATEGORY property is used as a free form
>>>      'tagging' field.  [...]
>>>
>>> In light of the discussion in the previous section about grouping without
>>> semantics, we might consider mentioning here that this is "tagging with
>>> semantics", just vaguely defined ones, as opposed to the REFID-type
>>> "tagging without semantics".
>> Added some text.
>>> Section 1.4
>>>
>>>      server-side management and stripping of inline data.  Clients may
>>>      choose to handle attachments differently from the LINK property as
>>>      they are often an integral part of the message - for example, the
>>>      agenda.
>>>
>>> (See the NITS entries as well, but) is it particularly noteworthy that
>>> clients might handle LINKs different from ATTACHments?  They are different
>>> properties after all -- why would someone assume equivalent treatment?
>> I'm not sure why - but it seems worth pointing out just in case. There's
>> some degree of similarity between a LINK and an ATTACHMENT - especially
>> if the attachment value is a uri.
>>
>> Also I think when I wrote this managed attachments wasn't an RFC. I have
>> the ref but the text is vague.
>>
>> ------------ OLD ----------
>>
>> The LINK property MUST NOT be treated as just another attachment.
>> The ATTACH property defined in [RFC5545] is being extended to handle
>> server-side management and stripping of inline data.  Clients may
>> choose to handle attachments differently from the LINK property as
>> they are often an integral part of the message - for example, the
>> agenda.
>>
>> For more information on managed attachments see [RFC8607]
>>
>> ------------ NEW ----------
>>
>> The LINK property MUST NOT be treated as just another attachment.
>> The ATTACH property defined in [RFC5545] has been extended by [RFC8607]
>> to handle server-side management and stripping of inline data and to
>> provide additional data about the attachment (size, filename etc).
>>
>> Additionally clients may choose to handle attachments differently
>> from the LINK property as attachments are often an integral part
>> of the message - for example, the agenda.
>> ---------------------------
>>
>>
>>> Section 1.5
>>>
>>>      To facilitate offline display the link type may identify important
>>>      pieces of data which should be downloaded in advance.
>>>
>>>      In general, the calendar entity should be self explanatory without
>>>      the need to download referenced meta-data such as a web page.
>>>
>>> I'm not sure how to relate these two statements.  It sounds like a "<X>
>>> may happen, but in general don't do <X>" scenario, in which case it might
>>> flow better to give the general advice first, followed by the specific
>>> scenario in which there is an exception to the general guidance.
>> It does read better - changed
>>> Section 2
>>>
>>>      The actual reference value can take three forms specified by the type
>>>      parameter
>>>
>>> Is this list fixed for eternity or do we want to give some guidance that
>>> implementations need to prepare for future extensibility?
>>>
>>>      REFERENCE:  In an XML environment it may be necessary to refer to a
>>>
>>> This qualification in the description ("XML environment") suggests that
>>> perhaps the name being used for these semantics is an overly generic name.
>> Having reread the text section has a number of problems, e.g - it's not
>> "type" - it's the VALUE parameter. I think the title is wrong - this is
>> really about LINK property references so I'll change the title. Also
>> LINK has no default type and - to get ahead a little - in the LINK
>> definition VALUE-TEXT should be VALUE=UID.
>>
>> Your point about REFERENCE stands - I'll change it to XML-REFERENCE
>> elsewhere.
>>
>> I'll update the text and get a new version out.
>>
>> I've added some text to point out that UID references may need updating
>> on import and export.
>>
>>> Section 4
>>>
>>> We do have some text here about the GAP parameter that specifies the lead
>>> or lag time here, but then we go on to describe the various RELTYPE values
>>> in language like "cannot start until", "can only be completed after",
>>> etc., that is very absolute and does not acknowledge the potential for a
>>> GAP.  I think it would be helpful to reframe as that we are making a
>>> comparison between the indicated pair of events, and that the relationship
>>> between when the events occur is affected by the GAP interval.
>>> (Also, the text in this section on the GAP parameter doesn't give a great
>>> sense that the lead time would be a negative gap.)
>> Rather than reframe I'll add a note and a couple of examples of the
>> relationships with a GAP parameter.
>>> Section 5
>>>
>>>      RELTYPE=FIRST:  Indicates that the referenced calendar component is
>>>         the first in a series the referenced calendar component is part
>>>         of.
>>>
>>> I wonder if we want to explicitly contrast this with PARENT and provide an
>>> example of them being different.
>> I added this to allow jscalendar and icalendar events to be mapped.
>> Unfortunately I omitted to add RELTYPE=NEXT.
>>
>> I'll add the NEXT relation,
>>
>> The difference is that PARENT, CHILD and SIBLING relationships are about
>> hierarchy and FIRST, NEXT about order. I can add some text.
>>
>>> Section 6.1
>>>
>>>         In addition to the values defined here any value defined in
>>>         [RFC8288] may be used.  However these uses SHOULD be documented in
>>>         an RFC updating both [RFC5545] and [RFC8288]
>>>
>>> In some sense this feels a little like saying "the current document SHOULD
>>> have documented this stuff, but we didn't".  Is there anything useful to
>>> say about why this documenting can't be done now?  (Oh, I guess there are
>>> rather more values in the registry than I remembered, though the number of
>>> them actually defined in RFC 8288 might be zero?)
>> In response to other comments I cut this back to
>>
>> In addition to the value defined here any link relation
>> in the link registry established by [RFC8288],
>> or new link relations, may be used.
>>
>> but I'm a little uneasy about that. Maybe add the text ", but should be
>> documented in an RFC" at the end?
>>
>> It seems to me that we should at least document how these are to be used
>> in a calendaring environment.
> Maybe something like "It is expected that link relation types seeing
> significant usage in calendaring will have the calendaring usage described
> in an RFC"?  There are probably a lot of different approaches here, and I
> have no strong sense of which would be "best".
>
>>> Section 8.1
>>>
>>>      Conformance:  This property can be specified zero or more times in
>>>         any iCalendar component.
>>>      [...]
>>>         Within the "VEVENT", "VTODO", or "VJOURNAL" calendar components,
>>>         more than one formal category can be specified by using multiple
>>>         CONCEPT properties.
>>>
>>> Are these two statements fully compatible?  The former seems to give
>>> leeway to put multiple CONCEPT properties in any iCalendar component, but
>>> the latter seems to limit to the "more than one" ability to just the three
>>> named components.
>> I suspect that's something that remained after a copy/paste exercise.
>> The paragraph naming components should go - it just repeats what was
>> written before it.
>>> Section 8.2
>>>
>>> I am not entirely sure when the TEXT value type would be used.  If it's
>>> just there to allow for certain types of extensibility, that might be
>>> worth stating specifically.
>> That should have been VALUE=UID.
>>>         There is no mapping for [RFC8288] "title*", "anchor", "rev" or
>>>         "media".
>>>
>>> Maybe "there is currently no mapping"?  Or is the definition of such
>>> mapping prevented for some technical reason(s)?
>> They don't work or have meaning in the calendar context
>>> Section 9
>>>
>>>      This specification updates the RELATED-TO property defined in
>>>      Section 3.8.4.5 of [RFC5545].
>>>
>>> I'd consider adding a note that "the contents of Section 9.1 below replace
>>> Section 3.8.4.5 of [RFC5545]" to clarify the relationship between the two
>>> sections.
>> Done
>>> Section 9.1
>>>
>>>         By default, the property value points to another calendar
>>>         component that has a PARENT relationship to the referencing
>>>         object.  The "RELTYPE" property parameter is used to either
>>>         explicitly state the default PARENT relationship type to the
>>>         referenced calendar component or to override the default PARENT
>>>         relationship type and specify either a CHILD or SIBLING
>>>         relationship or a temporal relationship.
>>>
>>> We allow some new relationship types that are neither CHILD/SIBLING nor
>>> temporal (e.g., CONCEPT, REFID).  Should those get some text in the
>>> description too?  (I am not sure if DEPENDS-ON and FIRST should get lumped
>>> in as "temporal", either -- we don't do so in §5.)
>> I'll change that to
>>
>>         relationship type and specify some other relationship.
>>
>> I'll add some text for the other types.
>>>         The FINISHTOSTART, FINISHTOFINISH, STARTTOFINISH or STARTTOSTART
>>>         relationships define temporal relationships as specified in the
>>>         reltype parameter definition.
>>>
>>> Likewise here, it seems some more discussion is in order for the other
>>> relationship types.
>>>
>>>         Changes to a calendar component referenced by this property can
>>>         have an implicit impact on the related calendar component.  For
>>>
>>> Do we want to say anything about changes to a component referenced by this
>>> property using any of the new relationship types?
>> I've added a paragraph to explicitly mention the possibility of broken
>> links.
>>> Section 10
>>>
>>> Many thanks for referencing the URI security considerations and calling
>>> out "reliability and consistency"!
>>>
>>> We might say that the security considerations of RFC 5545 continue to
>>> apply.
>>>
>>> There are perhaps some privacy considerations if a user uses the new
>>> CATEGORY or REFID functionality to expose information about a related
>>> group of events, but it's a little hard to concoct a scenario where this
>>> information gets to an attacker other than the calendar server, which is
>>> already in something of a trusted position with respect to the user.
>>>
>>> Are there any considerations about the new linking and relation operators
>>> exposing information to other users about calendar components that they're
>>> not authorized to access?
>>>
>>> Very large "gap" parameters seem likely to lead to unexpected behavior.
>>>
>>> NITS
>>>
>>> Section 1.4
>>>
>>>      iCalendar component.  This new LINK property is closely aligned to
>>>      the LINK header defined in [RFC8288]
>>>
>>> There are probably some pedantic distinctions that could be made here
>>> relating to how RFC 8288 defines the generic concept of Web Linking (which
>>> is not intrinsically tied to HTTP), as well as its expression in an HTTP
>>> header *field*.
>> How about:
>>
>> component. This new LINK property is closely aligned to
>> RFC8288  which defines the generic concept
>> of Web Linking as well as its expression in the HTTP LINK header
>> field.
> Looks good, thanks.
>
>>> Also, please put a full stop at the end of the sentence.
>> Done
>>>      The ATTACH property defined in [RFC5545] is being extended to handle
>>>      server-side management and stripping of inline data.  [...]
>>>
>>> It's hard (at least on first read) to tell if this sentence is referring
>>> to the RFC 8607 work or something being done by this document.
>> ...is being extended in other specifications to handle
>>>                                                            Clients may
>>>      choose to handle attachments differently from the LINK property as
>>>      they are often an integral part of the message - for example, the
>>>      agenda.
>>>
>>> Please clarify whether "they" refers to attachments of the LINK property.
>>> The singular/plural signal implies it's "attachments" but we could
>>> probably be more clear to the reader.
>>>
>>> Also, two hyphens for an em dash.
>> replaced "they" with "attachments"
>>> Section 4
>>>
>>>      This section defines the usual temporal relationships for use with
>>>
>>> What's the context behind "usual"?  I don't remember (but may have missed)
>>> a previous reference giving context for temporal relationships.
>> Maybe I should remove "the usual". Looking at things liek
>>
>> https://protect2.fireeye.com/v1/url?k=31323334-501d5122-313273af-454445555731-c6425350d8ccb11e&q=1&e=5275fae3-e0fe-41ff-b644-6d633dd1d924&u=https%3A%2F%2Fproject-management.info%2Fpdm-precedence-diagramming-method%2F
>>
>> these relationships are almost universal - but I don't know what could
>> be considered a standard. That organization does appear to create
>> standards but I don't know I can go much beyond saying "widely used".
>>
>>>      RELTYPE=FINISHTOFINISH:  Task-B can only be completed after Task-A is
>>>         finished.  The related tasks may run in parallel before
>>>         completion.
>>>
>>>         For example, if the goal is to prepare a meal, we start the
>>>         potatoes, then the meat then the peas but they should all be
>>>         cooked at the same time.
>>>
>>> This doesn't look like a great example for "finish-to-finish", as we
>>> prefer all things to finish at the same time, rather than with a strict
>>> ordering between end times.
>> In finish-to-finish we're trying to express that one task cannot be
>> completed until another task has completed. Both may run concurrently
>> but it's the end of both that matters. Perhaps this is a better example?
>>
>> ------------ New --------------
>>
>> For example, in the development of two related pieces of software, e.g.
>> the api and the implementation, the design of the implementation (B)
>> cannot be completed until the design of the api (A) has been completed.
>>
>> -------------------------------
> I like it better, at least.
>
>>> Section 6.1
>>>
>>>      Registration:  These relation types are registered in [RFC8288]
>>>
>>> I think we mean '''registered in the "Link Relation Types" registry
>>> established by [RFC8288]'''.
>> Yes
>>> Section 8.2
>>>
>>>      Property Parameters:  The VALUE parameter is required.  Non-standard,
>>>         reference type or format type parameters can also be specified on
>>>         this property.  The LABEL parameter is defined in [RFC7986]
>>>
>>> I'm not sure I see what in the ABNF would correspond to the "reference
>>> type" parameters.  The closest seems to be the linkrelparam, but that's a
>>> relation type, not a reference type.
>> I think it should be
>>
>>      Property Parameters:  The VALUE parameter is required.
>>            Non-standard, link relation type,
>>            format type, label and language parameters can also be
>>            specified on this property. The LABEL parameter
>>            is defined in [RFC7986]
> Ok, that does seem easier for me to make sense of.
>
> Thanks again for all the good stuff here,
>
> Ben
>
>>> Section 9.1
>>>
>>> Looking at the diff from RFC 5545, we should title case "Property Name"
>>> and "Value Type".  We also have changed the order in which we list the
>>> options for property parameters but that seems unimportant.
>> OK
>>> Section 11.1
>>>
>>>      The following iCalendar property names have been added to the
>>>      iCalendar Properties Registry defined in Section 8.3.2 of [RFC5545]
>>>
>>> Pedantically, RELATED-TO is not "added" but is rather covered by the "has
>>> added a reference to this document where ... have been updated by this
>>> document".  So if we were to keep this phrasing we might make two tables,
>>> or we could try to fudge the wording here a bit.
>>>
>>>
>>>
>>> _______________________________________________
>>> calsify mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/calsify