Hi,

I've been investigating the SNOMED ECL ANTLR grammar for use with a language other than Java. Whilst experimenting with the ANTLR tooling, I've observed that the following rules may not be reachable in the grammar:

• conjunctionrefinementset
• disjunctionrefinementset

Link to rules

The subrefinement rule appears to be greedy in the eclrefinement rule they are used in. All conjunction / disjunction combinations I've tried seem to just use the conjunctionattributeset/disjunctionattributeset rules.

Example below using the latest ANTLR grammar in ANTLR Lab:

ECL:

<  404684003 |Clinical finding| :
          363698007 |Finding site|  = <<  39057004 |Pulmonary valve structure|  AND
          116676008 |Associated morphology|  = <<  415582006 |Stenosis|  AND
          42752001 |Due to|  = <<  445238008 |Malignant carcinoid tumor|

Parse Hierarchy:

Are there scenarios that do use these two rules or could the grammar be simplified?

Hi,

I've observed some different behaviour when validating / executing ECL in the International SNOMED CT Browser / Snowstorm vs using the Grammar with ANTLR Lab or an ANTLR Parser targeting a different language e.g. C#.

If I execute an ECL query with some erroneous text after a valid expressionconstraint rule snowstorm correctly reports that there is some unexpected input e.g.

<< 404684003 |Clinical finding (finding)|  - Some invalid text : 363698007 |Finding site| = 89187006 |Airway structure (body structure)|

The error I get back is Syntax error at line 1, character 43: unexpected character '-' expecting {, '\u0009', '\u000A', '\u000D', ' ', '/'}

However validating the same ECL statement in ANTLR Lab the parser recognises << 404684003 as a valid expression constraint and ignores the remainder of the statement.

I've found that adding a new top level rule including an EOF reference fixes this behaviour and the parser in ANTLR Lab correctly provides the same error.

grammar ECL;

ecl : expressionconstraint EOF;
expressionconstraint : .... 

I have noticed that the snomed-ecl-parser repo uses an older version of the ANTLR Runtime (4.5.3), I'm wondering if there has been a behaviour change in the more recent versions (e.g. 4.13.1) of the ANTLR Runtime causing this difference?

However I did find this guidance on Starting Rules and EOF on the ANTLR repo, highlighting the same issue. Should there be a new top level rule in the grammar with an EOF marker to ensure the whole input document / ECL statement is read?