W3C home > Mailing lists > Public > public-automotive@w3.org > January 2020

Re: VSS Layers introduction

From: <Daniel.DW.Wilms@bmw.de>
Date: Wed, 22 Jan 2020 13:32:02 +0000
To: <ulfbjorkengren@geotab.com>
CC: <gandersson@genivi.org>, <mfeuer1@jaguarlandrover.com>, <rudolf.streif@ibeeto.com>, <public-automotive@w3.org>
Message-ID: <63F446DF-7D40-4390-A153-D1274DBDE31E@bmw.de>
> The new instantiation feature complicates how layers best are implemented in a tool.
The include complicates it in the same way. Even with the old structure you would have in the majority of the cases the additional information for the leaf. The old structure is even worse, because you’d have to specify throughout the path, where it splits where to find the correct layer-file. With the instances in the leaf only 😉

> The instantiation as such is very likely to be modified through layering. How did you handle that, @Daniel.DW.Wilms<mailto:daniel.dw.wilms@bmw.de>?
The deployment file is taken as input to the parser, as you suggest later in the mail.


> The solution I see is to have two tools (heard that before;?), one that takes the existing vspec files+instantiation data, and generate
> two files -
>  One that contains the complete, instantiated tree, including all metadata from the vspec files,
>  One that contains the complete, instantiated tree, excluding  all node metadata from the vspec files.

> The empty skeleton tree can then be populated with layers information, and then both files are fed to another tool that merges the
> two trees, and transforms it to other format(s).

I don’t see necessarily two kinds of tools, just another output of the parser. Like, instead of ‘make csv’ take ‘make layer-skeleton’.

Beste Grüße / Best regards,
Daniel

---
BMW Technology Office Israel
Daniel Wilms
Research Engineer

phone: +972 54 34 20 806
mail: daniel.dw.wilms@bmwgroup.com<mailto:daniel.dw.wilms@bmwgroup.com>

postal address:
BMW Technology Office Israel Ltd
121 Menachem Begin Road
LABS Mailbox numbers: 93-95
Tel Aviv
Israel 8067318



From: Ulf Bjorkengren <ulfbjorkengren@geotab.com>
Date: Wednesday, 22 January 2020 at 2:24 PM
To: "Wilms Daniel, LT-7-IL" <Daniel.DW.Wilms@bmw.de>
Cc: Gunnar Andersson <gandersson@genivi.org>, Magnus Feuer <mfeuer1@jaguarlandrover.com>, Rudolf J Streif <rudolf.streif@ibeeto.com>, public-automotive <public-automotive@w3.org>
Subject: Re: VSS Layers introduction

The new instantiation feature complicates how layers best are implemented in a tool. I agree that it must be possible to use the layering functionality for individual instantiated leaves.
The instantiation as such is very likely to be modified through layering. How did you handle that, @Daniel.DW.Wilms<mailto:daniel.dw.wilms@bmw.de>?

The solution I see is to have two tools (heard that before;?), one that takes the existing vspec files+instantiation data, and generate two files -
  One that contains the complete, instantiated tree, including all metadata from the vspec files,
  One that contains the complete, instantiated tree, excluding  all node metadata from the vspec files.

The empty skeleton tree can then be populated with layers information, and then both files are fed to another tool that merges the two trees, and transforms it to other format(s).

On Wed, Jan 22, 2020 at 12:10 PM <Daniel.DW.Wilms@bmw.de<mailto:Daniel.DW.Wilms@bmw.de>> wrote:
>> "a top level directory, where we can add the deployment files"
> For better or worse, the vspec interpretation depends on the directory structure, so I proposed the same for extensions.

Just to clarify: when I talk about extensions I understand adding signals or entire subtrees to the existing standard tree. There I would see the private branch as the right thing to use, which is already some kind of layer we have.

> Do you use a different way to define the node path in the extension files?  My default was to aim for consistency first and
> a minimal amount of extra new rules/exceptions.  It also ensures that files are close to the files they modify.  It seems
> straight forward to me...

I had in mind, that the tooling supports creating a skeleton of the layer-file from the vspec files, which you can modify then  for your specific deployment of the tree.


> Is a single deployment file enough, you ask?  I dont know without understanding its content...  I do not immediately see
> how it solves the variability/composability problem, but I could be wrong.

Theoretically I’m on your side and as said, we did it first like you are proposing. But there are problems, especially with the includes and now with the instances. This results in the fact, that the vspec doesn’t resemble the outcoming tree structure one to one to avoid duplication. Which makes a lot of sense. But for the layers, you want to have different meta data for each leaf, like mapping the correct CAN signal to a specific leaf. At the end, you don’t want to add the meta data to the more generic vspec file structure, but to the resulting tree/leaf. Therefore, I’d vote for a clear separation of the layers and the vspec files.

Whereas, additional terms (subtree or leaf) in the taxonomy should be done through the private branch, which needs to be extended to support merge, etc.

Daniel

---
BMW Technology Office Israel
Daniel Wilms
Research Engineer

phone: +972 54 34 20 806
mail: daniel.dw.wilms@bmwgroup.com<mailto:daniel.dw.wilms@bmwgroup.com>

postal address:
BMW Technology Office Israel Ltd
121 Menachem Begin Road
LABS Mailbox numbers: 93-95
Tel Aviv
Israel 8067318



From: Gunnar Andersson <gandersson@genivi.org<mailto:gandersson@genivi.org>>
Date: Tuesday, 21 January 2020 at 8:23 AM
To: "Wilms Daniel, LT-7-IL" <Daniel.DW.Wilms@bmw.de<mailto:Daniel.DW.Wilms@bmw.de>>, "mfeuer1@jaguarlandrover.com<mailto:mfeuer1@jaguarlandrover.com>" <mfeuer1@jaguarlandrover.com<mailto:mfeuer1@jaguarlandrover.com>>, "rudolf.streif@ibeeto.com<mailto:rudolf.streif@ibeeto.com>" <rudolf.streif@ibeeto.com<mailto:rudolf.streif@ibeeto.com>>, "ulfbjorkengren@geotab.com<mailto:ulfbjorkengren@geotab.com>" <ulfbjorkengren@geotab.com<mailto:ulfbjorkengren@geotab.com>>
Cc: "public-automotive@w3.org<mailto:public-automotive@w3.org>" <public-automotive@w3.org<mailto:public-automotive@w3.org>>
Subject: Re: VSS Layers introduction

Thanks everyone for input, and sorry for the top post. Bad mail client, and limited time.

It is perhaps not "meant" to have multiple layers, it is just a possibility.  I am uncertain if you misunderstood or if I am reading your questions wrong.  But generally, yes, I am proposing separating different concerns in different files, especially if the variability is such that composability of those layers models the reality in a good way.  And I expect more variability than only "market".

The purpose of mentioning markets was just an example where we have variability and therefore it is (one) reason to put the variable part in a layer, so that you can keep the constant part (the standard vspec for example) unchanged.  We agree on the premise so far?

In my view the market specific info is a layer, but with multiple instances (one per market).
You exchange the version of that layer when it needs changing (a file version per market, or whatever other thing is being modeled.  They are instances of the concept being modeled).

If this does not fit market specific info then maybe we look at other ways there.  I still think the layer concept is the generic tool for this type of stuff, but we could drill into market variations to see if it works.

Complex?  I think (hope) I am only proposing a generic optional tool/approach here and that any complexity stems from the reality that is modeled, primarily.

Is a single deployment file enough, you ask?  I dont know without understanding its content...  I do not immediately see how it solves the variability/composability problem, but I could be wrong.

> "a top level directory, where we can add the deployment files"

For better or worse, the vspec interpretation depends on the directory structure, so I proposed the same for extensions.  Do you use a different way to define the node path in the extension files?  My default was to aim for consistency first and a minimal amount of extra new rules/exceptions.  It also ensures that files are close to the files they modify.  It seems straight forward to me...

Anyway, the input from Rudi explains the implementation is also quite flexible in how you interpret files.  I think the key is to agree on convention or best practice and describe that too, for real world modeling scenarios.  I mean current python tools in the repo is one thing and the formal specification (rules/recommendations/conventions) of what VSS *is*, is another.  The latter is possibly more important to agree on.  Because tools will be written and extended over time in any case.
--
Gunnar Andersson <gandersson@genivi.org<mailto:gandersson@genivi.org>>
Development Lead, GENIVI Alliance
________________________________
From: Daniel.DW.Wilms@bmw.de<mailto:Daniel.DW.Wilms@bmw.de>
Sent: Monday, January 20, 2020 21:24
To: mfeuer1@jaguarlandrover.com<mailto:mfeuer1@jaguarlandrover.com>; rudolf.streif@ibeeto.com<mailto:rudolf.streif@ibeeto.com>; ulfbjorkengren@geotab.com<mailto:ulfbjorkengren@geotab.com>
Cc: gandersson@genivi.org<mailto:gandersson@genivi.org>; public-automotive@w3.org<mailto:public-automotive@w3.org>
Subject: Re: VSS Layers introduction

Hi,

First of all thanks @Gunnar for sharing the presentation. I really like the concept. I don’t have much to add, as it maps almost one to one how we do it. I would have two remarks from our experience:

  1.  From slide 7 I understood, that it’s meant to have multiple layers on top of each other. But that gets pretty complex then, doesn’t it? I agree that you have to support different markets etc., but can’t you put those specifics into one layer? We called the same concept “deployment”, where we collect the additional meta-data for one specific deployment and it contains everything, which is needed for the system where the tree gets deployed. Maybe that’s enough to keep the complexity manageable?
  2.  Slide 9: You mention that the directory/file structure should be the same, just a different suffix. We started doing it the same way, but noticed that it quickly was hard to maintain. For the spec it works as it’s more stable, but on a deployment/layer level, which constantly changes and needs adaptation, it was pretty hard to keep the overview. So we decided to go for a top level directory, where we can add the deployment files and where one file contains the additional meta information. We saw that it was much easier to handle.

>> Adding or overriding metadata in nodes should be straight forward. Removing metadata in nodes needs a little more design thinking, maybe with a separate "remove-
>> tool" and layer files?
> Removing nodes from the default tree through an extension tree is a use case that has not yet been discussed.
>> What to me seems more complicated is to add or remove nodes, is that something that should be supported?

I think removing nodes is not an issue. The moment you don’t add any additional meta information to a leaf it’s removed in the deployment of the tree. If a subtree doesn’t have any leaf it’s removed as well, etc.

Removing meta-data would change the spec quite heavily. Should it be supported? Further, fields like datatype, unit, etc. shouldn’t be overwritten, because that’s the core of the specification. Maybe instances would be some natural candidate to be overwritten.

> Adding nodes is supported. The whole scheme is actually pretty simple if a node already exists when an extension/layer file is processed then the fields get updated (or added for that matter if a field does not yet exist e.g. CAN DB field). If a node does not yet exist it will be added.

Adding nodes shouldn’t be supported through layers. Wouldn’t it be done through the spec and not the meta information? If not public, it would be the private branch. Otherwise it kills reusability. You would have to know each and every layer to get the full picture of the tree. Further, you end up with different naming, data-types, etc. Even on a private level you don’t want to go there.

Daniel

---
BMW Technology Office Israel
Daniel Wilms
Research Engineer

phone: +972 54 34 20 806
mail: daniel.dw.wilms@bmwgroup.com<mailto:daniel.dw.wilms@bmwgroup.com>

postal address:
BMW Technology Office Israel Ltd
121 Menachem Begin Road
LABS Mailbox numbers: 93-95
Tel Aviv
Israel 8067318



From: Magnus Feuer <mfeuer1@jaguarlandrover.com<mailto:mfeuer1@jaguarlandrover.com>>
Date: Monday, 20 January 2020 at 10:18 PM
To: Rudolf J Streif <rudolf.streif@ibeeto.com<mailto:rudolf.streif@ibeeto.com>>, Ulf Bjorkengren <ulfbjorkengren@geotab.com<mailto:ulfbjorkengren@geotab.com>>
Cc: Gunnar Andersson <gandersson@genivi.org<mailto:gandersson@genivi.org>>, W3C Public Automotive <public-automotive@w3.org<mailto:public-automotive@w3.org>>
Subject: Re: VSS Layers introduction
Resent-From: <public-automotive@w3.org<mailto:public-automotive@w3.org>>
Resent-Date: Monday, 20 January 2020 at 10:17 PM

Late to the party, as per usual.

We do very much need layers. As Ulf and Rudi pointed out, we need to think about removing both metadata and whole branches, for example when a single instance of something needs to be replaced by an enumerated list of instances.

A somewhat silly example is if we have multiple IVI systems, which necessitates the following change:

...Infotainment.[subtree] -> ...Infotainment.1.[subtree], ...Infotainment.2.[subtree]...

In this case the existing "...Infotainment.[subtree]" needs to be deleted since it would interfere with the enumerated list of IVIs.

One (somewhat over-engineered) solution would be to add metadata (attribute) to the overlay file's the branch, specifying how the overlay should be applied:

#
# I am an overlay file that modifies an existing tree.
#
- Vehicle.Cabin.Infotainment:
  type: branch
  overlay: replace # or one of  extend, modify, modify_and_extend.

- Vehicle.Cabin.Infotainment.1:
  type: branch
...

The overlay attribute can be one of:
extend
The overlay extends an existing branch in the signal spec, but does not modify existing signals. If an existing signal is re-defined by the overlay, an error is thrown.
Metadata can be added to existing signals.

modify
The overlay modifies existing branches and signals in the original spec. If a signal (or branch) does not exist in the original, an error is thrown.
Metadata can be added to existing signals.

modify_and_extend [default]
The overlay can both modify existing branches and signals, and create new one.

replace
The entire branch is cleared from the original specification, and only the structure defined in the overlay will be used.

Would that work?

/Magnus F


-------------------
System Architect Manager
Jaguar Land Rover

Email: mfeuer1@jaguarlandrover.com<mailto:mfeuer1@jaguarlandrover.com>
Mobile: +1 949 294 7871

Error! Filename not specified.
Jaguar Land Rover North America, LLC
1450 NW 18th Ave, Portland, OR 97209
-------------------
Business Details:
Jaguar Land Rover Limited
Registered Office: Abbey Road, Whitley, Coventry CV3 4LF
Registered in England No: 1672070

This e-mail and any attachments contain confidential information for a specific individual and purpose.  The information is private and privileged and intended solely for the use of the individual to whom it is addressed.  If you are not the intended recipient, please e-mail us immediately.  We apologise for any inconvenience caused but you are hereby notified that any disclosure, copying or distribution or the taking of any action in reliance on the information contained herein is strictly prohibited.

This e-mail does not constitute an order for goods or services unless accompanied by an official purchase order.



________________________________
From: Rudolf J Streif
Sent: Monday, January 20, 2020 10:44
To: Ulf Bjorkengren
Cc: Gunnar Andersson; W3C Public Automotive
Subject: Re: VSS Layers introduction


Hi Ulf,

> I like that the VSS layers model as I understood it takes as input "layer files" with the same filename, but different extension, located in the same directory as the
> vspec file that they are to update.

Actually that is not how it currently works. The overlay files can have different names and are in different directories. What is important is the branch name of the nodes and that they files are included at the right position in the tree. That means, yes, you can have a file with a different name (including extension) in the same directory and include it there or you can have a full directory tree starting at the root. And of course anything between these two "extremes".

> With this model we will have the existing tool(s) that transform the vspec YAML format to other formats (JSON, csv, etc), and new tool(s) that use the YAML layers to
> modify the YAML vspec files.

The converter tools already handle that.

> Adding or overriding metadata in nodes should be straight forward. Removing metadata in nodes needs a little more design thinking, maybe with a separate "remove-
> tool" and layer files?

Removing nodes from the default tree through an extension tree is a use case that has not yet been discussed.

> What to me seems more complicated is to add or remove nodes, is that something that should be supported?

Adding nodes is supported. The whole scheme is actually pretty simple if a node already exists when an extension/layer file is processed then the fields get updated (or added for that matter if a field does not yet exist e.g. CAN DB field). If a node does not yet exist it will be added.

:rjs
On 1/20/20 5:36 AM, Ulf Bjorkengren wrote:
I like that the VSS layers model as I understood it takes as input "layer files" with the same filename, but different extension, located in the same directory as the vspec file that they are to update.
With this model we will have the existing tool(s) that transform the vspec YAML format to other formats (JSON, csv, etc), and new tool(s) that use the YAML layers to modify the YAML vspec files.
Adding or overriding metadata in nodes should be straight forward. Removing metadata in nodes needs a little more design thinking, maybe with a separate "remove-tool" and layer files?
What to me seems more complicated is to add or remove nodes, is that something that should be supported?

On Fri, Jan 17, 2020 at 7:31 PM Rudolf J Streif <rudolf.streif@ibeeto.com<mailto:rudolf.streif@ibeeto.com>> wrote:
Actually this statement is not quite correct:

1) Potential for *any number of* layers. (As of now the private branch
acts as a *one* additional "layer" that can override and add data, as
you now described)


It's not limited to a branch called 'private'. What branch is merged and
where is defined by an include in the toplevel
VehicleSignalSpecification.vspec. Now of course this begs the question
what happens if there are multiple includes that modify/amend the same
nodes when there are conflicts? The answer is: "the last one wins" as
the includes are processed in order of appearance in
VehicleSignalSpecification.vspec. That essentially constitutes an
implicit priority.

Layer priority is probably something that you want to add to the slide
"Additional design challenges".

:rjs

On 1/16/20 9:52 PM, Gunnar Andersson wrote:
>
> Thanks for pointing that out, Rudi.
>
> (continued...)
>
> On Wed, 2020-01-15 at 09:32 -0800, Rudolf J Streif wrote:
>> Functionality of appending and/or modifying a node from one branch
>> with
>> a node from another branch with the same signal path is already
>> available in VSS.
> [trimmed]
>
>> https://github.com/GENIVI/vehicle_signal_specification/commit/518dfbeb2a434c9c55ec27e06c84f6e8caf941bd#diff-4a931512ce65bdc9ca6808adf92d8783

>>
> That's great and shows that this thinking was there from the beginning.
> That commit indeed provides an implementation that is a starting point
> for the behavior that this proposal looks for.
>
>>> The commit message states:
> [trimmed]
>
>> Now that does not cover everything from Gunnar's slides (in
>> particular
>> there is no wildcard support) but it's a starting point.
> Yes.  And a few other differences that come to mind.  These are things
> that might be added on top of the current model, *if* the layers idea
> were adopted:
>
> 1) Potential for *any number of* layers. (As of now the private branch
> acts as a *one* additional "layer" that can override and add data, as
> you now described)
>
> 2) A proposal of layers as parallell files in the same directory tree
> -- with different file suffix --  (it would complement the current
> model that places /private as a parallel directory subtree)
>
> 3) The presentation hints at some additional higher level concepts that
> would allow defining metadata about the layers themselves.
> and as a corrollary it also suggests that we are effectively defining
> some varying file types.  I.e. they might not be formally .vspec" files
> but something else.   (The current /private "layer" does not explicitly
> make a distinction, and basically assumes ".vspec" files to be there).
> I'm not sure how important that is but it seemed useful to me.
>
>
> Best Regards,
> - Gunnar
>
> P.S. Since we are now talking about VSS specifics, I expect the details
> of this will be sorted out using GitHub issues, but if it is in the
> interest of the W3C participants then of course by all means feel free
> to continue the discussion here.
>
>
>
>
>
>
> It's a great start however.
>
>> :rjs
>>
>> On 1/14/20 12:26 PM, Gunnar Andersson wrote:
>>> As requested today, here is a link to the introduction/rationale
>>> slides
>>> for "VSS Layers".
>>>
>>> A small caveat, this presentation is a work in progress and
>>> describes
>>> the rationale on a very fundamental level.  The idea is simple of
>>> course, and very generic, because it is useful/needed in a whole
>>> bunch
>>> of areas, not only for access control even if that that was one of
>>> the
>>> driving areas.
>>>
>>> There of course needs to be some work to tie this into the W3C
>>> protocol
>>> specifics by means of examples etc.  We concluded today that it is
>>> basically in line with what is needed in Ulf's recent
>>> implementation
>>> work as well as previous similar things that Daniel mentioned had
>>> been
>>> tried in his organization in the past, so hopefully this is useful,
>>> even if it does not today provide all the concrete examples in the
>>> W3C
>>> protocol context.
>>>
>>> Happy to have the discussion about comparing this to previous
>>> similar
>>> ideas and to define more of the details.
>>>
>>> Thanks,
>>> - Gunnar
>>>
>>> [1]
>>> https://at.projects.genivi.org/wiki/pages/viewpageattachments.action?pageId=40402952&sortBy=date&highlight=VSS+composable+layers+-+draft+20200114.pdf&&preview=/40402952/47711024/VSS%20composable%20layers%20-%20draft%2020200114.pdf

>>>
>>> or shorter:
>>> https://bit.ly/2TnrV05

>>>
>
--
-----
Rudolf J Streif
CEO/CTO ibeeto
+1.855.442.3386 x700


--
Ulf Bjorkengren

Geotab

Senior Connectivity Strategist | Ph. D.

Mobile

+45 53562142

Visit

www.geotab.com<https://www.geotab.com/>



--

-----

Rudolf J Streif

CEO/CTO ibeeto

+1.855.442.3386 x700


--
Ulf Bjorkengren

Geotab

Senior Connectivity Strategist | Ph. D.

Mobile

+45 53562142

Visit

www.geotab.com<https://www.geotab.com/>



Received on Wednesday, 22 January 2020 13:32:44 UTC

This archive was generated by hypermail 2.4.0 : Wednesday, 22 January 2020 13:32:45 UTC