Re: (Lost in the noise perhaps - so asking again) - Is a trailing slash 'better' than a trailing hash for vocabs namespace IRIs?

Hi Pierre-Antoine,

Thanks so much for engaging in the discussion - I really, really appreciate
it!

You bring up some great points, but in my view, and with a bit more
clarification from my side, I think they all seem to actually be
reinforcing the argument for slashes (just as a *preference* when creating
new vocabs)!

In the general case, when you encounter an IRI of the form http://ex.co/x/Y,
you can not assume that http://ex.co/x/ will contain the definition of
http://ex.co/x/Y together with other related terms.

Yep, absolutely, I totally agree - this is the wild-wild-west of the World
Wide Web after all, so yeah, regardless of slashes or hashes, we can never
assume anything at all when dereferencing any IRI anywhere.
And so yeah, nobody should ever make the assumption you point out here (it
never even occurred to me that anyone would!). As you say below, that's why
`rdfs:isDefinedBy` is so useful (and should be a general Best Practice for
vocabs anyway, I'd say).

For this you need,

a) the server to provide an affordance in the description of
http://ex.co/x/Y pointing to http://ex.co/x/  (e.g. by using
rdfs:isDefinedBy)

Yep - totally agree. Which is why I always highly recommend (as a separate,
but related, general Best Practice) that all vocab terms should always
provide an `rdfs:isDefinedBy` triple regardless. (Thankfully that term's
local name (i.e., 'isDefinedBy') is very intuitively self-explanatory!)

b) the client to understand and follow that affordance.

Well, yeah, kinda, but only *if* that client *wants* to be able to take
advantage of that extremely handy and helpful little affordance to
follow-its-nose. And given the client needs to understand RDF (to some
minimal degree at least) in the first place to be working with RDF vocabs
at all, that doesn't seem like a problem to me, or any kind of an issue at
all.

In other words, all I'm stating is that if we *prefered* slashes, then any
client wishing to understand what any individual IRI *is* can simply
deference that IRI (i.e., isn't that just the first principle of Linked
Data really!). And they should always (in my opinion) be able to expect to
get back *only* a representation of whatever that IRI represents. So *if*
the IRI they dereferenced happened to be an individual vocab term, then
(only with slash-based vocabs) they'd correctly get back data on just that
one vocab term.

(I'd consider it merely a Best Practice that they might *also* be able to
expect a `rdfs:isDefinedBy` link to the overall vocab within which that
single vocab term is defined - but they only need to understand and/or
follow that link if they ever wanted to *also* discover information on the
containing is-defined-by vocab.)
So in other words, I only see goodness here, and a simple consistent
expression of Linked Data first principles (i.e., all 'things of interest'
should have uniquely *dereferencable* IRIs, and you can choose to
follow-your-nose to 'more info' if you want to, and you understand the
predicates leading the way).

c) the description at http://ex.co/x/ to include some information about any
term (e.g. http://ex.co/x/Z) in contains stating "there is nothing more to
know about this term" (e.g. by using rdfs:isDefinedBy again)

I don't quite follow this point - perhaps you could elaborate a little, or
provide some sample Turtle...? (For example, I would expect the description
at http://ex.co/x/ (assuming that to be a vocab namespace IRI) to indeed
contain *all* of the information about the vocab itself (like it's
versioning info, preferred prefix, creation date, etc.), and *all* the
information about *all* of the terms that that vocab contains/defines -
i.e., exactly as QUDT do today when you click on their namespace IRI:
https://qudt.org/schema/qudt/ (although they only seem to provide Turtle,
and not a content-negotiable full HTML representation that I'd prefer to
see them provide from my browser (e.g., DPV does provide a lovely 'complete
vocab' that is a content-negotiable (i.e., HTML or Turtle) representation
when you dereference it's namespace IRI today: http://www.w3.org/ns/dpv#)).


d) the client to understand that statement and refrain from fetching
http://ex.co/x/Z later on

I didn't follow the above point, so this one loses me too, but (I think)
this comes down to clients needing to know (regardless of slash or hash)
the basic difference between an `owl:Ontology` and an `rdf:Class` or
`rdfs:Property` (i.e., between 'a full vocab' and 'a term in a vocab') in
the responses they get from servers when they are dereferencing
vocab-related IRIs anyway. I don't think the issues around caching are
majorly affected by the slash/hash choice - but perhaps I'm missing your
real point here...


So you don't get "the best of both world" as automatically as you suggest.

Oh yeah, absolutely - so we agree again. I should have been clearer perhaps
- I certainly didn't mean to imply that getting the-best-of-both-worlds was
in any way 'automatic' at all. As I said later in my post, getting both
will require more work on the server side, or from tooling.
All I'm trying to emphasize is that slashes provide *a means* to get
the-best-of-both-worlds, whereas with hashes the best servers can ever
offer to clients (regardless of the needs or wants of those clients) is to
return information on all terms in the entire vocab (since, by HTTP design,
the server will never receive the hash fragment in any HTTP request), and
so you can never, ever offer any client *the option* to just retrieve a
single vocab term's information and nothing else *if that's what the client
wants/needs/prefers*.

Terms of a vocabulary/ontology rarely make sense in isolation. So arguably,
serving the entire vocabulary provides you with enough context to
understand/use the term appropriately.

Well, I wouldn't agree with that at all. I think QUDT's CurrencyUnit (
https://qudt.org/schema/qudt/CurrencyUnit) is a great example of where it
makes perfect sense (i.e., all I want to know is what QUDT *mean* by a
'CurrencyUnit'). And surely no-one would argue that Schema.org should
switch from its current slash to use hash instead, because terms like
Person (https://schema.org/Person) need context from the entire 2,500 terms
defined in Schema.org as a whole.

But I do certainly agree with your point that individual terms should
indeed provide enough context to understand/use the term - but I think all
that context should be provided *in isolation* within the vocab's
definition of that term itself, and should certainly not require
downloading the entire vocabulary - i.e., examples of precisely that are
`rdfs:isDefinedBy`, `rdfs:domain`, `schema:rangeIncludes`, `rdfs:seeAlso`,
`skos:related`, `skos:narrower`, etc.
Now, given that much of that 'term-specific context' would actually be
IRIs, it should then be up to the client to decide if it now wishes to
dereference each of those individual links with multiple HTTP requests, or
if it actually wishes to now download the entire vocabulary in one HTTP
request (again, only slashes offer all clients the choice and flexibility
for them to decide between those options for themselves).

And then you get "bombarded with a huge document"...

Yep, but again my point is that only with slash do clients get at least the
option, or the ability, or the possibility, to *choose for themselves*
whether they get bombarded with the entire vocab in one HTTP request or not.


So doesn't that demonstrate my whole point - i..e, that with slashes I can
get the best of both worlds

I don't think so. They are different trade-offs between providing targeted
content vs. reducing the number of HTTP queries, and between working with
dumb clients and/or dumb servers vs. requiring more coordination between
them  (e.g. providing and following rdfs:isDefinedBy links).

Well to emphasize my point, with slashes I *can* get
the-best-of-both-worlds, with hashes I *can't*.
Yep, for sure there are tradeoffs, and indeed implementing the full set of
options (with full conneg, and providing/generating individual
term-specific static HTML pages, etc.) definitely requires more server-side
work/tooling. But I'd argue that adopting slash still doesn't *require*
that any of that work be done at all, for example, if all you have to work
with are dumb servers - i.e. your dumb server can just continue as always,
serving up the full vocab information for all requests using the exact same
single static page it uses today with hash, by simply using a single URL
rewrite rule to rewrite http://ex.co/x/Z to http://ex.co/x#Z
<http://ex.co/x/Z>. Sure, that breaks the client expectation somewhat
(i.e., "I only asked for info on term Z, and you gave me info on all the
vocab terms - but at least you provided the HTML anchor so that my browser
jumped down automatically to the info for term Z!") - but that's a
worst-case scenario (i.e., a scenario that may have been forced on you due
to only having dumb servers and no server-side tooling), and yet it's still
no worse than what you get today with hashes (i.e., it *is* what you get
today with hashes).

And that's why I posit that slashes are simply 'more correct' (i.e., since
*only* slashes can ever allow servers to always know exactly,
unambiguously, what a requesting client is really looking for

I don't by that. The server can never know exactly nor unambiguously what
the intent of the client is, nor should it (separation of concerns).

Sure, of course :) ! So let me re-phrase my point - only slashes provide
the means for the server to *see* the *full/complete IRI* that a client may
wish to de-reference. In other words, with hashes, by HTTP design, the
client *must* strip off the hash fragment (if any) before putting the HTTP
request on the wire - hence the server can't ever distinguish between a
client asking for info on a single term (e.g., GET http://ex.co/x#Z
<http://ex.co/x/Z>) or a client wishing for info on the entire vocab (e.g.,
GET http://ex.co/x# <http://ex.co/x/Z> or just GET http://ex.co/x
<http://ex.co/x/Z>).


Cant' help but cite the priority of consituencies remininded in
https://www.w3.org/TR/design-principles/#priority-of-constituencies

Yep, exactly (we agree again!) - but for me, this is precisely an argument
for slashes - i.e., hashes restrict what clients can possibly get back from
a server (i.e., they'll always get the full vocab information back),
whereas slashes at least provide *the potential* for servers to offer
clients more flexibility and choice (i.e., info just on individual terms,
*or* info on the full vocab).
So surely giving clients *more* choice (with slashes), not less (with
hashes), is putting their needs first (since we can't possibly ever know
up-front, for any vocab, the 'needs' of all potential users (i.e., the
entire user base of the Web) for vocabs we publish, right!?

Also, in a distributed setting such as the web, you can not assume that all
other parties will always do the right thing™.

Again, I totally agree (who wouldn't!).
But all I'm looking for here is this community's opinion on whether we can
offer a clear, *single*, *preference* for the creators of new RDF
vocabularies going forward. I think we can, and that *preference* should be
using slashes (i.e., not a requirement, or a mandate, or something anyone
can ever force people to do). I just think the current state of guidance in
the Linked Data community is too loose and therefore off-putting for
newbies - i.e., "You can do either, there are pro's and con's, but it
doesn't really matter much, so you can just decide for yourself". Newbies
don't want to have to 'decide for themselves' if they can help it when
learning new technology - and so they'll just continue the current practice
of cutting-and-pasting what they see as most prevalent out there today
(e.g., nearly all the W3C vocab examples today), which will most likely
mean repeating the 'mistake' of using hashes, and thereby 'hurting' the
longer-term options for client/user software that may wish to have the
ability (at some future stage perhaps) to be able to choose for themselves
between term-specific or full-vocab lookups.

Thanks again Pierre-Antoine for pushing me to think this through even more
thoroughly - I hope it's been somewhat useful for you (and others?) to
ponder on too :)

Pat.


*Pat McBennett*, Technical Architect

Contact  | patm@inrupt.com

Connect | WebID <http://pmcb55.inrupt.net/profile/card#me>, GitHub
<https://github.com/pmcb55>

Explore  | www.inrupt.com





On Fri, Oct 7, 2022 at 8:07 AM Pierre-Antoine Champin <pierre-antoine@w3.org>
wrote:

> On 07/10/2022 01:49, Pat McBennett wrote:
>
> Hi Martynas,
>
> Thanks for the feedback!
>
> But I think any vocabulary can just as easily support that same caching
> benefit with slash-based vocab namespace IRIs too, *without* having
> to require an initial HTTP request for *each* term - i.e., by simply
> returning the entire vocab on namespace IRI lookups.
>
> In the general case, when you encounter an IRI of the form
> http://ex.co/x/Y, you can not assume that http://ex.co/x/ will contain
> the definition of http://ex.co/x/Y together with other related terms. For
> this you need,
>
> a) the server to provide an affordance in the description of
> http://ex.co/x/Y pointing to http://ex.co/x/
> b) the client to understand and follow that affordance (e.g. by using
> rdfs:isDefinedBy)
> c) the description at http://ex.co/x/ to include some information about
> any term (e.g. http://ex.co/x/Z) in contains stating "there is nothing
> more to know about this term" (e.g. by using rdfs:isDefinedBy again)
> d) the client to understand that statement and refrain from fetching
> http://ex.co/x/Z later on
>
> So you don't get "the best of both world" as automatically as you suggest.
>
>
> I think QUDT is a really nice, simple example that very easily
> demonstrates exactly this today. It has a slash namespace IRI, and if I
> only ever request info on individual single vocab terms (e.g., try clicking
> now on `https://qudt.org/schema/qudt/CurrencyUnit`) then yes, I'd
> encounter that 'HTTP request per lookup' you suggest (but I'd be getting
> precisely what I asked for each time!).
>
> Terms of a vocabulary/ontology rarely make sense in isolation. So
> arguably, serving the entire vocabulary provides you with enough context to
> understand/use the term appropriately.
>
>
> But I can just as easily avoid that scenario today too by simply
> requesting the vocab's namespace IRI instead - e.g., try it right now by
> just clicking on `https://qudt.org/schema/qudt`
> <https://qudt.org/schema/qudt>. See - you get back the entire vocab
> containing all the vocab terms in a single HTTP response, which can be
> cached and keyed on that one namespace IRI (exactly as you would if they'd
> used a hash instead).
>
> And then you get "bombarded with a huge document", to quote one of your
> arguments against hash-IRIs. Seems to me that you get the worst of both
> worlds here: I had to perform two HTTP queries (one on CurrencyUnit, got
> get the link to the whole vocab, and one on the vocab) instead of one (with
> hash IRIs), and I still end up with a huge ontology. (yes, playing devil's
> advocate here a little)
>
> (I'm not familiar with Jena's OntDocumentManager, but I'm sure its
> caching code could easily be extended to take advantage of servers that
> choose to server up slash-based vocabularies as QUDT demonstrates is so
> feasible today.)
> So doesn't that demonstrate my whole point - i..e, that with slashes I can
> get the best of both worlds
>
> I don't think so. They are different trade-offs between providing targeted
> content vs. reducing the number of HTTP queries, and between working with
> dumb clients and/or dumb servers vs. requiring more coordination between
> them  (e.g. providing and following rdfs:isDefinedBy links).
>
> (i.e., precise term-specific HTTP responses if I want them, *and* the
> entire vocab in a single HTTP response if I want that too)? Using a hash
> completely locks me out, forever, of being able to achieve those lovely
> clean term-specific responses.
> And that's why I posit that slashes are simply 'more correct' (i.e., since
> *only* slashes can ever allow servers to always know exactly,
> unambiguously, what a requesting client is really looking for
>
> I don't by that. The server can never know exactly nor unambiguously what
> the intent of the client is, nor should it (separation of concerns
> <https://en.wikipedia.org/wiki/Separation_of_concerns>).
>
> (i.e., a term-specific response, or an entire vocab response)), and it
> does so without losing any of the benefits of using a hash. (I do, by the
> way, totally appreciate that servers choosing to work as the QUDT servers
> do today might require a bit more server-side work. But my whole point is
> to ask this community which option they consider "more technically correct
> today and forever", and not "which option is easier for servers or vocab
> creators/hosters/editors/publishers today in the absence of any tooling
> support".
>
> Cant' help but cite the priority of consituencies remininded in
> https://www.w3.org/TR/design-principles/#priority-of-constituencies
>
> "User needs come before the needs of web page authors, which come before
> the needs of user agent implementors, which come before the needs of
> specification writers, which come before theoretical purity."
>
> Don't get me wrong, I get the point of thinking beyond the limitation of
> current tools. That's a valuable exercise. But practicality does also
> matter.
>
> Also, in a distributed setting such as the web, you can not assume that
> all other parties will always do the right thing™.
>
>   my 2€
>
>   pa
>
> In other words, I think that QUDT-server-like behaviour can be provided
> easily by tooling, which I'd personally be very happy to work on
> contributing :) !).
> Cheers,
> Pat.
>
> *Pat McBennett*, Technical Architect
>
> Contact  | patm@inrupt.com
>
> Connect | WebID <http://pmcb55.inrupt.net/profile/card#me>, GitHub
> <https://github.com/pmcb55>
>
> Explore  | www.inrupt.com
>
>
>
>
> On Thu, Oct 6, 2022 at 3:44 PM Martynas Jusevičius <martynas@atomgraph.com>
> wrote:
>
>> Hi Pat,
>>
>> For one thing, hash URIs are easier to cache because there is only one
>> document URL. After the initial HTTP request the whole document can be
>> cached with its URL as the key. All following term lookups (whose URIs
>> start with that URL) will hit the cached document.
>> Slash URIs will require an initial HTTP request for *each* term and will
>> result in a cache entry per term.
>>
>> This is based on my experience with Jena's OntDocumentManager.
>>
>> Martynas
>> atomgraph.com
>>
>>
>> On Thu, Oct 6, 2022 at 4:15 PM Pat McBennett <patm@inrupt.com> wrote:
>>
>>> So (I think!) I know all the pro's and con's of using either a trailing
>>> slash or a trailing hash for vocab namespace IRIs. Basically it boils down
>>> to hashes meaning you'll always get info on all the terms in a vocabulary,
>>> even if you only want info for one specific term, whereas using a slash
>>> means I can always get just the info for any specific, individual term I
>>> request.
>>>
>>> Note: using slashes provides the ability to get the best of both worlds
>>> - i.e., small responses when explicitly asking for info on just one term,
>>> but if you want info for all the terms in one HTTP response, then just
>>> serve up that complete vocab response when the base namespace IRI itself is
>>> dereferenced.
>>>
>>> Here's a nice simple illustration of the basic difference:
>>> - Slash: QUDT's 'CurrencyUnit' term (i.e., click on '
>>> https://qudt.org/schema/qudt/CurrencyUnit') and you get a nice clean,
>>> concise, and precise set of info on just the one term you asked for -
>>> lovely!
>>>
>>> - Hash: DPV's 'JointDataControllers' (i.e., click on '
>>> https://w3id.org/dpv#JointDataControllers') and you get bombarded with
>>> a huge document, with a daunting Table of Contents on the left, and info on
>>> hundreds of other terms that I didn't ask for, and so had no interest in
>>> whatsoever (don't get me wrong - this is fantastically detailed and
>>> potentially very useful information, but it's simply not what I asked for!).
>>>
>>> So based on the greater flexibility and future-proofing of using slash
>>> (i.e., it offers the best of both worlds, whereas hash is forever limited),
>>> I've become firmly of the opinion that slashes are just 'better' that
>>> hashes, and in fact are simply 'more correct' (i.e., all IRIs should be
>>> uniquely dereferencable).
>>>
>>> I also think the distinction is critically important when creating
>>> vocabularies intended for widespread and long-lasting use (such as the DPV
>>> vocab above). For throw-away or pet projects, sure, it doesn't really
>>> matter (yet even then, I still think slashes are the 'more correct' option).
>>>
>>> I know that the convention from the W3C has tended to be to use hashes,
>>> but I think in hindsight that was a mistake, and that the advice from the
>>> Semantic Web community as a whole should now be to adopt slashes
>>> consistently for all new vocabularies. (And it's not like using slash has
>>> no precedent - major 'authoritative' vocabs like QUDT, Schema.org, gist,
>>> SOSA, SSN, (even the venerable FOAF!) all use slash).
>>>
>>> I'd love to hear this group's thoughts. (For reference, I did ask the
>>> gist community if they recorded their discussions around their decision (in
>>> 2019) to formally switch gist from hash to slash (here
>>> <https://github.com/semanticarts/gist/issues/725>), but it seems they
>>> weren't recorded, and I've also raised the issue with the DPV group
>>> directly too (here <https://github.com/w3c/dpv/issues/53>)).
>>>
>>> Cheers,
>>>
>>> Pat.
>>>
>>> *Pat McBennett*, Technical Architect
>>>
>>> Contact  | patm@inrupt.com
>>>
>>> Connect | WebID <http://pmcb55.inrupt.net/profile/card#me>, GitHub
>>> <https://github.com/pmcb55>
>>>
>>> Explore  | www.inrupt.com
>>>
>>>
>>>
>>> This e-mail, and any attachments thereto, is intended only for use by
>>> the addressee(s) named herein and may contain legally privileged,
>>> confidential and/or proprietary information. If you are not the intended
>>> recipient of this e-mail (or the person responsible for delivering this
>>> document to the intended recipient), please do not disseminate, distribute,
>>> print or copy this e-mail, or any attachment thereto. If you have received
>>> this e-mail in error, please respond to the individual sending the message,
>>> and permanently delete the email.
>>
>>
> This e-mail, and any attachments thereto, is intended only for use by the
> addressee(s) named herein and may contain legally privileged, confidential
> and/or proprietary information. If you are not the intended recipient of
> this e-mail (or the person responsible for delivering this document to the
> intended recipient), please do not disseminate, distribute, print or copy
> this e-mail, or any attachment thereto. If you have received this e-mail in
> error, please respond to the individual sending the message, and
> permanently delete the email.
>
>

-- 
This e-mail, and any attachments thereto, is intended only for use by the 
addressee(s) named herein and may contain legally privileged, confidential 
and/or proprietary information. If you are not the intended recipient of 
this e-mail (or the person responsible for delivering this document to the 
intended recipient), please do not disseminate, distribute, print or copy 
this e-mail, or any attachment thereto. If you have received this e-mail in 
error, please respond to the individual sending the message, and 
permanently delete the email.

Received on Saturday, 8 October 2022 01:10:33 UTC