The current requirements for Derived Property conformance (including the changes accepted in https://issues.omg.org/browse/SYSMOAS_-102 ) are too narrow.

• For the Query service, the text only focuses on the derived properties being computed and included in the results. However, it should also clarify whether the query can refer to (i.e. filter by) the computed values of derived features.
• For the Element Navigation Service, the text only mandates that the data structure returned at commit creation should have the derived properties filled in. However, ideally, derived features should also be filled in when requesting element contents for reading, without having updated them; this guarantee is omitted in the current text.
  • See the discussion at https://issues.omg.org/browse/SYSMOAS_-102?focusedCommentId=27339&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-27339 on why this is NOT already implied by the current text

The Query model in PIM, specifically PrimitiveConstraint entity has value attribute data type as String [ 1..* ].

However, the PrimitiveConstraint's value attribute in REST/HTTP PSM JSON schema is not typed as Array, causing inconsistency between PIM and PSM models.

The value data type in both Query PIM and PSM should be an array.

Similarly to ProjectUsage, ExternalRelationship inherits from Data, hence PIM level operation definitions should be consistent for both. At the moment PIM level ExternalRelationshipService group is lacking operations for ExternalRelationship creation and deletion when compared to ProjectUsageServices, hence:

1. PIM level ExternalRelationshipService group should have createExternalRelationship and deleteExternalRelationship operations added
2. REST/HTTP PSM level operation to service mappings should provide steps on how createExternalRelationship and deleteExternalRelationship are realized through the createCommit operation

Summary of rationale for SYSMOAS_-83 (Read-only API providers)

The issue SYSMOAS_-83 (https://issues.omg.org/browse/SYSMOAS_-83) put forth a good suggestion on Conformance levels. A summary of that suggestion and its purpose is stated below.

API and Services Conformances, as specified in section 2 of the specification, are directly based on individual services and require that all operations (get/create/update/delete) for the given service be implemented for a service provider to claim conformance to that service. For example, ProjectService Conformance states the following.

ProjectService Conformance - A Service Provider must implement all the operations in the ProjectService, and demonstrate that the implementation successfully passes all the ProjectService Conformance Test Cases (see A.1) and Cross-Cutting Conformance Test Cases (see A.7).

A digital engineering tool, such as a Requirements management tool or a Product Lifecycle Management (PLM) platform, may expose some of its data using services defined in this standard (Systems Modeling API and Services 1.0) using only the get operations. For example, a Requirements management tool may implement the get operations in ProjectService and ElementNavigationService to fetch requirement projects and requirement structures. Similarly, a PLM platform may implement the same to fetch hardware projects and part BOMs. However, these implementations would be non-conformant to both the ProjectService and ElementNavigationService since they would only implement the get operations and not the create/update/delete (data mutation operations).

Summary of work items for SYSMOAS_-83 (Read-only API providers)
SYSMOAS_-83 also suggested work items (included in its description) that can be summarized as below.

1. Identify get operations from mutation operations (create/update/delete) by using the 'query' UML modifier on those operations.

2. Define Read-level and Read-Write Conformance levels for each service so that non-SysML digital engineering tools that make their data available using get operations can claim Read-level conformance.

Problem with proposal SYSMOAS_-86 for SYSMOAS_-83

The proposal SYSMOAS_-86 for issue SYSMOAS_-83 took an incorrect step with major change impacts that was NOT part of the core work items in SYSMOAS_-83 and also summarized above.

Instead of limiting the change to splitting the Conformance levels for each service in section 2 for Read-Only and Read-Write Conformance, the proposal also split the individial PIM services. A new supertype service was proposed for each PIM service with get operations, e.g. ProjectServiceReadOnly with get operations, inherited by ProjectService that will also include the mutation (create/update/delete) operations.

This proposal of splitting the PIM services was not necessary to achieve the goal of SYSMOAS_-83 and has a major change impact on this specification, as described below:

1. It was not necessary to split the PIM services. Splitting of the Conformance levels in Section 2 and stating that Read-Only Conformance can be achieved by implementing the get operations (also marked by 'query' UML modifier) in each service was sufficient to achieve the original aim. Only the conformance test suite in Appendix A would have to be refactored. No PIM level services, or PSM services/endpoints, or PIM -> PSM mappings would have been affected.

2. Splitting each PIM service also impacts the mapping from PIM -> PSM for both REST/HTTP PSM and OSLC PSM. The mapping tables have to be redone, and the PSM level service groups may have to be modified. For example, GET endpoints in the REST/HTTP PSM may require to be separated from POST/PUT/DELETE. The extent of this change is significant and requires a lot more time to investigate and plan even before affecting the change. Given that FTF2 was closing in August when the proposal was approved, it was infeasible from the start in addition to being unncessary. This is true even with the new Nov 2024 deadline for submission.

3. In the future, as this standard and the service providers evolve, we may need to define new conformance levels or refactor existing conformance levels. Changes in the conformance levels should not require us to rearchitect the design of the PIM services which then subsequently impacts the PSM services and the PIM -> PSM mappings.

4. It is also not recommended to name services with "read" or "write". This is done at the operation level. A service is a grouping of operations related to an entity, and should remain so.

Solution being proposed in this issue
This issue is being opened to correctly implement the core work items of SYSMOAS_-83 and to rollback the PIM service splitting step in proposal SYSMOAS_-86. A draft proposal will be forumated in the comment of this issue, and following a discussion in the FTF2 meeting, a formal proposal will be created an attached to this issue.

in the description of getCommitChange and diffCommits they reference an optional argument changeTypes, which doesn't appear in figure 10

says in the standard "Branches are immutable", no they are mutable... can change the head subsequently

The property doesn't seem to be composite in the API

We shall codify an additional invariant that ensures Commit timestamps increase monotonically, i.e. part commits have a strictly older timestamp.

Reasoning:

• this would make comparing and ordering Commits easier; given two commits from the history of a branch, it would take a single step (just comparing timestamps) to determine which one came earlier, without having to crawl through a potentially long chain of commits.
• there is little extra burden brought by the introduction of this constraint, as it is already how most use cases would work, with Commits being created one by one and timestamps being naturally assigned at the time of Commit creation; this additional invariant would only really constraint edge cases, such as when a large volume of history is imported from an external format, and timestamps are synthetically assigned en masse

Follow-up from https://issues.omg.org/browse/SYSMOAS_-83 and specifically the comment https://issues.omg.org/browse/SYSMOAS_-86?focusedCommentId=26800&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-26800 by Mr. Hans Peter de Koning

Two of the Service interfaces, namely ElementNavigationService and ExternalRelationshipService, are entirely read only, and thus are not affected by the resolution of https://issues.omg.org/browse/SYSMOAS_-83 - they work with read only API providers as is.

However, as Mr. Hans Peter de Koning argues, we should treat them the same way as all the other Service interfaces. Even though this is not useful at the moment, there are two good reasons to extend the change:

• Consistency among parts of the API: by having different parts of the API resemble each other in a systematic way and share common logic or naming conventions, Clients and Providers will both find it easier to (a) implement communication through the API and (b) communicate about required or supplied conformance to API services
• Future proofing: in case future editions of the standard introduce read-write operations to these two API services, there will already be a place ready for them in the standard, both in the normative machine-readable models and the language of the text.

Right now, the standard assumes repositories that provide any of the listed API services to implement all operations of these services, both the read only ones and the mutating ones. I believe we should have at least vocabulary in the standard to describe and recognize the kind of partial compliance with a given API Service where only the read-only operations of the Service are actually provided.

*Reasoning.* In order for a service to claim SysML compliance, it must implement some combination of the 6 existing API services defined in the standard. Most of which prescribe mutation operations such as createProject(), etc. but especially and primarily createCommit(), that make little sense to implement if the single source of truth is a data format other than SysML (or the textual format of SysML), and the SysML format is merely a derived view of the data.

*Example.* An off-the-shelf requirement managements system might not be willing to try and implement arbitrary SysML model mutation commands (especially pertaining to non-requirement stuff), and on the other hand creating requirements entries might require data fields that are not mandatorily written via the SysML API - at any rate this is a bad match.

*Core proposal.* It would make a lot of sense in these use cases if the standard talked about an officially recognized way for these services to claim that they implement some Services of the SysML API - but only the read-only operations. This does not mean that the projects would be immutable, just that they are to be modified through a means other than the SysML API.

*Clarification.* This request is not about user-specific access permissions. Even in the current version of the standard, any SysML v2 repository may in principle be configured so that certain users are forbidden from executing operations that manipulate state, there will still be other users that are authorized to execute them. In contrast, the proposed change would introduce repositories where (certain types of) mutation of state is simply not accomplished through SysML v2 API, but through other means.

*Impact on testability.* Repositories that provide the read-only form of one of the Services should still comply with the semantics and constraints of the read-only operations exposed by the Service. However, since the state of such repositories is not (fully) controllable via the SysML v2 API, this conformance relationship may not be possible to test via the Conformance Test Suite as is. Providers of such repository technologies are expected to demonstrate conformance via modified test cases, adapted to set up and influence repository state by means appropriate for the specific technology, with effects matching the original conformance test as closely as possible.

*Work items.* I propose to elaborate in detail the following changes to the standard:

• Section 2 "Conformance" should be expanded to (a) discuss Service Providers that only expose the read-only fragment of a Service, (b) provide vocabulary to discuss the degree of conformance by Service Providers, and (c) discuss how the conformance of such Service Providers is to be demonstrated
• Section 7.2 "API Services" should identify the subset of operations that are read only, featured both on Tables and on Diagrams (perhaps using the 'query' UML modifier)
• Normative Machine Readable Documents should be updated to reflect in operation metadata which of the API operations are read only; in particular, the PIM XMI should include the UML modifier 'query' for each such operation.

From Tim Weilkiens <tim.weilkiens@oose.de>

I got the following question from someone who does some experiments with the API:

Here is an excerpt from the SysML v2 API specification:

> For creating new Data, CommitRequest.change should include a DataVersion where DataVersion.payload includes the Data being created and DataVersion.identity is not specified.

It says that DataVersion.identity must not be specified when creating new data.

However, how can one commit new data with multiple elements that reference each other, if we cannot provide IDs to the elements that will allow each of them to be referenced from other elements via its ID?

As a side note, it has been noticed that the Juypter notebook support for SysMLv2 in the SysML release does specify the “identity” during the creation of new data, when it publishes a model via “%publish …”. This has been verified by using an HTTP sniffer to sniff the POST command that was send

Note: the PIM spec also says something similar

> (1) Creating Data - Commit.change should include a DataVersion record with DataVersion.payload populated with the Data being created. DataVersion.identity is not provided, thereby indicating that a new DataIdentity needs to be created in the new commit.

Within ProjectDataVersioningService PIM-level service specifications createCommit service Element deletion rules state the following:

When a DataIdentity is deleted in a commit, all its versions (DataVersion) are also deleted, and any references from other DataIdentity are also removed to maintain data integrity. In addition, for Element Data (KerML), deletion of an Element must also result in deletion of incoming Relationships.

This seems to be model-breaking, especially in cases of ownership-based Relationships (e.g. OwningMembership, FeatureMembership). Example:

Namespace – (OwningMembership Relationship) --> Package packA – (OwningMembership Relationship) --> PortDefinition portA.

When deleting Package packA, according to the rule described, an incoming OwningMembership Relationship should be deleted as well. However, the outgoing OwningMembership Relationship pointing to PortDefinition portA would remain to exist in the model leaving portA floating without any owner (OwningMembership Relationship has no source, it has been deleted).

Simply deleting outgoing Relationships in addition to incoming ones would also not help in this case, as the PortDefinition portA would also remain without any owner in the model.

We need to take a closer look into various Relationship kinds and their semantics in order not to break the model when manipulating via API services.

ProjectUsages are Data and so need to use the DataIdentity/DataVersion mechanism but none of this is mentioned in the descriptions of the functions. It also doesn't aay whether you can update a project usage to reference a different commit or

The argument to the function updateQuery is a query - I don't understand what this means - it also seems to be different than how updateProject works.

What happens if the resolution is that an identity is deleted?

e.g. getElements( project : Project, commit : Commit ) : Element [0..*]

commit references project so you don't need both.

createProject( name : String, description : String [0..1] ) : Project

how do we create other properties of Project inherited from Record?

property is a property of Data (or its realizations)
value is a list of primitive objects, such as String, Boolean, Integer, Double, and UUID

so why are they both typed by string?

The spec says that "select is a list of properties of Data (or its realizations)" so why is it typed by string and not something stronger?

The spec says "and represents the relationship between a
KerML Element [KerML] in a provider tool or repository to ExternalData in another tool or repository." This sounds very vague given that the neither the notions repository or tool have official definitions.

The spec also says "ExternalData may be a KerML Element or a non-KerML Element." I don't see how this is significant given that access is via IRI anyway.

the spec says "• owningProject is the project that owns the given branch" - this is a redefinition of owningProject on commitReference but I don't understand why it is needed.

The spec says "usage is the set of Project Usage records representing all other Projects being used by the given Project (Project Usage.usedProject)" but I can't see this in the xmi

the spec for project says "identifiedData is a derived attribute that is the set of Data Identity records corresponding to the Data contained in the project" but this doesn't seem to exist in xmi.

The spec says "Data Version.id is set to a new, randomly generated UUID for the specific Data Version record.". This seems a bit superfluous and probably is overspecified.

createAt and deleteAt attributes don't exist in the XMI.

The spec says "Each realization of Data must implement the getId() operation that provides a valid UUID." but this doesn't seem to be the case

https://www.omg.org/spec/SystemsModelingAPI/20240801/OSLC-Domain-Model.zip, needs updated vocabulary and constraints files for KerML and SysML:

• KerML Shapes-shapes.xml
• KerML Vocabulary-vocab.xml
• SysML Shapes-shapes.xml
• SysML Vocabulary-vocab.xml

The vocabulary terms updates include:

• Use updated SysML.ecore and KerML.ecore source files for generating the vocab files.
• Add the ontology element to the vocab .ttl file.
• Generate Turtle instead of RDF/XML
• Add Apache License Version 2.0 comment headers to the .ttl files.
• Add OASIS copyright notice.
• KerML and SysML v2 Element subclass oslc_am:Resource to link SysML elements to other lifecycle resources (e.g., requirements, change requests, test cases, etc.)
• Add rdfs:isDefinedBy oalc_sysmlv2 and rdfs:label to each vocabulary terms
• Translate the EEnum enumerations to RDF, e.g., FeatureDirectionKind
  See [Defining Enumerations](file:///Users/jamsden/Developer/OSLC/oslc-op/oslc-specs/specs/core/core-vocab.html#enumerations). An EEnum would be an rdfs:Class (with label, comment and definedBy). Its enumeration literals have rdf:type of the enumeration class, and just a label.
• Add SysML subclasses in the SysML vocabulary terms
• Use the SysML v2 non-versioned namespace for the vocabulary terms: @prefix oslc_sysmlv2: <http://www.omg.org/spec/sysmlv2/2.0#> .
• Set rdfs:comment in the vocab file to be the first paragraph in the eCore description with the markup stripped out.
• Fix ShapeChecker errors

The resource shape constraint updates include:

• Used updated SysML.ecore and KerML.ecore source files for generating shapes files.
• Add the ResourceShapeConstraints element to the shapes.ttl file
• Generate Turtle intead of RDF/XML
• Add Apache License Version 2.0 comment headers to the .ttl files.
• Add OASIS copyright notice.
• Add inherited oalc_am:Resource properties to the shapes for KerML and SysML v2 Element
• Update the versioned namespace for the resource shape constraints: @prefix : <https://www.omg.org/spec/SysML/shapes/20240801#> .
• Set dcterms:description in the shapes files to the first paragraph in the eCore description (retaining the markup).
• Separate the class name prefix from the attribute name with _ for the ResourceShapes
• Add cardinality, range and valueType to shape properties.
• Fixed ShapeChecker errors

the standard should specify how the commits (and previousCommit(s) ) are transverse to get all the elements

Two Branches can have the same head (Commit) … need multiplicity change
for example to start them off with the same set of elements

since DataIdentity is in the scope of the whole project, one can have multiple Branches where this DataIdentity is created in different Commits in different branches... and deleted in different Commits in different Branches...

so these cannot be associated with the DataIdentity... DataVersion's commit is where that information must be stored...

one could however have an operation that takes a project and branch and DataIdentity... and computes the Creation and Deletion Times from this

not sure one needs scope here... how would it work?... needs documentation in the spec about its use and an example

does the standard want to deal with locking on versioning systems?

Data Version record is associated with up to one (0..1) Data Identity record, this is wrong it is multiplicity of 1

need OCL to specify... previousCommits must be in the same project as the Commit

Data does not need to be further refined in the standard (i.e. no need for Relationship and operations like getRelationshipsByRelatedElements)... these could be there for operations that specifically deal with KerML models...
I would suggest if doing that there should be two sections to this standard or possibly two standards... one that deals with
versioning... the other that is specific to the serialized model

I have a reworked model at https://material.elparazim.com/SysMLv2API/

"usage is the set of Project Usage records representing all other Projects being used by the given Project"… this does not work, need a particular commit for this

What does project usage mean?... I am presuming this comes from Cameo kind of idea about ProjectUsage... but what in this spec does it mean and what can one surmise about it

currently in the spec a ProjectUsage refers to a project that is also local on the system (Server)...
there needs to be a way to reference a project (external) on another server...

I have a reworked model at https://material.elparazim.com/SysMLv2API/

Head on Branch should be multiplicity 0..1 ... when Branch is first created there is no commit

Projects may have multiple DataVersions… change Multiplicity

There should be a navigation between DataVersion to DataIdentity... presumably DataVersion knows about DataIdentity but not vice-versa

Two Tags should be able to be on the same Commit… need multiplicity change

Would suggest OCL for ProjectUsage and CommitReference constraints

suggestions in model at https://material.elparazim.com/SysMLv2API/

ProjectUsage needs a derived attribute “usedProject”

“created” attribute in Project, CommitReference, Commit should be

... these are not the only ones

I have a reworked model at https://material.elparazim.com/SysMLv2API/

there needs to be a statement in the standard concerning how things are serialized...

I have reworked the model here https://material.elparazim.com/SysMLv2API/ and used two stereotypes...
<<NS>> and <<SID>> which stand for "No Serialization" and "Serialize Id (only)"
I believe this leads to a full understand of what will come when one serializing something from
the APi (also examples are given for this serialization(s))

the standard for functions should be, e.g.

getExternalRelationships(projectId:UUID, branchId:UID[0..1],commitId:UID[0..1]):ExternalRelationship[0..*]

which means if a branchId is missing use the commitId, if commitId is missing then use the head of the branchId if it is there,
otherwise, use the default Branch and its head of the project given
many of the API operations need to be specified this way to be consistent

I have a reworked model at https://material.elparazim.com/SysMLv2API/

Most of the API is over specified… as an example, getExternalRelationship has that you need to give it a Project… but you do not need a Project, you only need a ProjectId… so for this one example…
getExternalRelationships(project:Project,commit:Commit):ExternalRelationship[0..*] should be
getExternalRelationships(projectId:UID,commitId:UID):ExternalRelationship[0..*]

this would follow the Interface Segregation Principle (“no client should be forced to depend on methods it does not use”)… most of the API is over specified in this way

I have a reworked model at https://material.elparazim.com/SysMLv2API/

specification and language need to be

... this will have the same kind of semantics as
Opaque/Expression or Behavior from UML (see fig 8.2 and its description in UML 2.5.1)

Need at least one example of a specification for an ExternalRelationship...

also need to define what it means if the specification is null (i.e. presume this means that one is referring to the entire element, so this is the way one specifies that an element in a model is really a remote element from another model)... this needs to be specified

need to explain in the standard why value is 1..* (presuming because of the "in" operator, but the semantics of that is not specified)

Fig. 7 orderBy should be

Fig. 7 queries (which should be a singular, query) is an associationend… whereas in fig. 5 queries is shown as an attribute (needs to be removed in fig. 5)

1. Fig. 7, Operator and JoinOperator need to be shown as enumerations, with literal compartments

1. Fig 6 visibility is not shown on all the attributes, needs to be consistent
2. Fig 7 visibility is not shown on all the attributes, needs to be consistent

Fig 5 Project, Commit, CommitReference all have created as a attribute, But many of the examples (and some of the text have timestamp).. needs to be consistent… I like “created” as it tells one what the timestamp is for

and deleted as the other timestamp name

would suggest using either MOF or UML to show diagrams and to employ ownership dots... note that KerML is specified this way

there are some multiplicities not shown... since there is no specification in the standard about what modeling language the diagram is in... do not know how to interpret these missing multiplicities... would suggest explicitly showing all multiplicities in the diagram

the models used in the standard have the association ends wrapped so you can't see them... suggest that they be unwrapped and
shown on one line

all the standard modeling languages use the same convention as well as most modeling style guides... and so does KerML...

all attributes and association ends are singular... the fact that they can be plural is taken care of by multiplicity

would suggest this convention be followed

IRI standard is referenced, but that does not give a precise definition of how to use it, especially if one wants to be interoperable among implementations... would suggest...

Standard ProjectI IRI:

://

/projects/

/commits/

the first is for Elements and the second is for externally referenced projects

resourceIdentifier is not needed in the record (although would be interesting)... statements about that in another issue... but remove alias, humanidentifier... and description... for example DataVersion and DataIdentity does not need a description... would suggest creating a NamedRecord where name and description are added into that and have Project, Commit, CommitReference, Query specialize that

it does not specify what language these models are in... UML, SysML, KerML, SysML2, MOF?...
it makes a difference in, e.g., interpreting the multiplicities... I would suggest doing the models in MOF and making everything explicit (e.g. multiplicities)

one needs Unique IDs (UID) but not UUID (Universally Unique ID) for most things in this standard,
and would suggest that Record be changed and that it would have

id : UID

also adding

here... and make Project (which is the only thing I can see that needs a UUID)

Project
id : UUID

there is nothing in this standard specifically about SysML or KerML... which is good... I would suggest that the Scope be rewritten and the standard be called "Versioning" or something like that... there is greater use of this standard inside modeling (i.e. legacy modeling language can use this as well) as well as outside of modeling (i.e. anything that can serialized can use this standard to have versions of that serialization)... so would suggest the Scope be changed to say

"The purpose of this standard is to specify the Versioning Application Programming Interface (API) and Services that provide standard services to access, navigate, and operate on serialized elements and in particular can be used for modeling languages."