Possible errors in the KEBNF grammars

[asmyers]

I really appreciate the KEBNF files in the repository! In working to get a functioning grammar from them I had to make a number of changes. I had claude go through my commit history and summarize each change, here they are:

1. Add missing FilterPackageImport rule (KerML KEBNF)

   The FilterPackage rule referenced FilterPackageImport, but this rule was never defined. Added the rule with both membership-import and namespace-import alternatives, matching the Xtext definition.

2. Add missing InvocationTypeMember rule (KerML KEBNF)

   The FunctionOperationExpression rule referenced InvocationTypeMember, but this rule was never defined. The Xtext grammar defines it as a trivial delegation to InstantiatedTypeMember. Added the missing rule.

3. Add AllocationDefinition to DefinitionElement (SysML KEBNF)

   AllocationDefinition was defined as a standalone rule but was missing from the DefinitionElement choice, so allocation def could never appear in a package or definition body. Added it as an alternative.

4. Add EndUsagePrefix alternative to OccurrenceUsagePrefix (SysML KEBNF)

   The KEBNF defined OccurrenceUsagePrefix using only BasicUsagePrefix, but the Xtext grammar shows it must also accept EndUsagePrefix (to support end occurrence usages like end port ...). Wrapped the body in a choice between EndUsagePrefix and the existing BasicUsagePrefix path.

5. Fix EntryTransitionMember reference: TargetSuccession to TransitionSuccession (SysML KEBNF)

   The EntryTransitionMember rule referenced TargetSuccession in the 'then' branch, but the correct rule per Xtext is TransitionSuccession. TargetSuccession is a different construct used in action successions.

6. Fix FramedConcernUsage reference: CalculationUsageDeclaration to ConstraintUsageDeclaration (SysML KEBNF)

   The FramedConcernUsage rule referenced CalculationUsageDeclaration, which does not exist. The Xtext grammar uses ConstraintUsageDeclaration, consistent with the pattern for concern/constraint usages.

7. Make 'standard' optional in LibraryPackage (SysML KEBNF + KerML KEBNF)

   The KEBNF used ( isStandard ?= 'standard' ) — the ?= operator means "set to true if present", implying the keyword should be optional, but the surrounding parentheses without ? made it a required group. Changed to ( isStandard ?= 'standard' )? in both files.

8. Add DefaultInterfaceEnd to StructureUsageElement (SysML KEBNF)

   Bare end usages (e.g., end myPort : PortType;) inside definition bodies require DefaultInterfaceEnd, but this alternative was missing from StructureUsageElement. The Xtext grammar includes it.

9. Add trailing dot to FlowEndSubsetting (SysML KEBNF)

   The KEBNF rule FlowEndSubsetting defined the qualified-name branch as referencedFeature = [QualifiedName] without a trailing ., making it impossible for the parser to distinguish the subsetting context from the flow feature member that follows. The Xtext grammar includes the . as a separator.

10. Fix SendNode declaration: ActionUsageDeclaration to ActionNodeUsageDeclaration (SysML KEBNF)

    SendNode used ActionUsageDeclaration? for its optional declaration prefix, but the Xtext grammar uses ActionNodeUsageDeclaration?. This parallels the pattern used by AcceptNode, AssignmentNode, and TerminateNode. The difference matters because ActionUsageDeclaration includes ValuePart? which creates ambiguity with the send payload.

11. Make 'assert' and 'not' optional in SatisfyRequirementUsage (SysML KEBNF)

    The KEBNF required 'assert' before 'satisfy' and made isNegated ?= 'not' non-optional (same ?= without outer ? issue as the LibraryPackage fix). The Xtext grammar makes both keywords optional.

12. Fix CaseBodyItem to use CalculationBodyItem instead of ActionBodyItem (SysML KEBNF)

    CaseBodyItem referenced ActionBodyItem as its base, but CaseDefinition extends CalculationDefinition in the metamodel, and calculations support ReturnParameterMember. The Xtext grammar uses CalculationBodyItem.

13. Add SysML ExpressionBody override using CalculationBodyPart (SysML KEBNF)

    Body expressions (e.g., { in ref w = ...}) inside SysML calculations and constraints could not parse because the KerML ExpressionBody rule uses FunctionBodyPart, which only allows KerML-level syntax inside the braces. The Xtext SysML grammar explicitly overrides ExpressionBody to use CalculationBodyPart. Added a SysML-specific override.

14. Add UsageExtensionKeyword* prefix to EnumeratedValue (SysML KEBNF)

    Enumerated values with prefix metadata annotations (e.g., #Security enum value1;) could not parse because EnumeratedValue did not allow UsageExtensionKeyword* before the optional 'enum' keyword. The Xtext grammar includes this prefix.

[hpdekoning]

Thanks, Andrew, for documenting these issues.

As a warning, the current state of these grammars is purely informative. There are quite a few more issues still. Work is in progress to systematically apply corrections as part of the updates for KerML v1.1 and SysML v2.1. I will add a README to the bnf folder describing the current state of the grammars.

[asmyers]

That would be great 👍
I'm working on a pipeline to mechanically transform the KEBNF into a tree-sitter grammar and I have something that parses the 309 sysml files returned by find SyML-V2-Release/ -iname "*.sysml" without error, but I haven't done a full correctness evaluation yet.
Any additional information on problems in the KEBNF would be appreciated so I can fold that in and create test cases for it.

[asmyers]

One last issue with the KerML grammar:

The KEBNF spec required FeatureDeclaration in Expression, BooleanExpression,
and Invariant rules, but the Xtext pilot wraps these with
ExpressionDeclaration (FeatureDeclaration? ValuePart?) making it optional.
Bare inv { expr } is valid KerML and used throughout the Kernel Library.

diff --git a/specs/KerML-textual-bnf.kebnf b/specs/KerML-textual-bnf.kebnf
index ea93371..e2e057a 100644
--- a/specs/KerML-textual-bnf.kebnf
+++ b/specs/KerML-textual-bnf.kebnf
@@ -900,7 +900,7 @@ ResultExpressionMember : ResultExpressionMembership =
 
 Expression =
     FeaturePrefix
-    'expr' FeatureDeclaration ValuePart?
+    'expr' FeatureDeclaration? ValuePart?
     FunctionBody
 
 // Clause 8.2.5.7.3 Predicates
@@ -913,13 +913,13 @@ Predicate =
 
 BooleanExpression =
     FeaturePrefix
-    'bool' FeatureDeclaration ValuePart?
+    'bool' FeatureDeclaration? ValuePart?
     FunctionBody
 
 Invariant =
     FeaturePrefix
     'inv' ( 'true' | isNegated ?= 'false' )?
-    FeatureDeclaration ValuePart?
+    FeatureDeclaration? ValuePart?
     FunctionBody


 // Clause 8.2.5.8 Expressions Concrete Syntax

[seidewitz]

@asmyers
These are not errors in the KEBNF files as such. They are errors in the normative grammars in the specification documents. The KEBNF files accurately reflect the current grammars per the specification.

Since they are part of the normative specification, issues with them need to be reported to OMG and handled though the specification revision process. Several of the ones in your list have already been report (sometimes more than once, e.g., KERML11-83, KERML11-113 and KERML11-179). These will be fixed in an upcoming beta release of the KerML 1.1 and SysML 2.1 specifications.

[seidewitz]

The KEBNF spec required FeatureDeclaration in Expression, BooleanExpression,
and Invariant rules, but the Xtext pilot wraps these with
ExpressionDeclaration (FeatureDeclaration? ValuePart?) making it optional.
Bare inv { expr } is valid KerML and used throughout the Kernel Library.

I have submitted this as issue KERML11-181.

Issues can be submitted at https://issues.omg.org/issues/create-new-issue. However, they need to be submitted against the specification documents, not the BNF files.

[antoineatstariongroup]

We do have similar findings while parsing KEBNF files. We would like to base our code-generation pipeline using grammar as source of truth.
Looking forward to see evolution there !