enforce matching terminals in grammar

[swamidass]

What is the correct way to ensure that a terminal is remembered and then matched later? The reason why is because I am trying to restrict LLM output to a SMILES string.

See here for a description: https://en.wikipedia.org/wiki/Simplified_Molecular_Input_Line_Entry_System

And here for a grammar that almost does it: https://depth-first.com/articles/2020/04/20/smiles-formal-grammar/

Here are some examples of valid strings. Note that the digit functions to close loops in a chemical structure. Whenever a specific digit appears, it must appear at least one more time. So these are valid:

C1CCC1
C1CC2CCC1CCC2
C2CCCC2
C3CCC3CCC2CCCC2
C1CCC1CCC1CCCC1

However, these are invalid:

C1CCC
C1CCCCC1CCC2
C2CCCC3
C3CCCCCC2CCCC
C1CCC1CCC1CCCC

Is there a way to do this with an EBNF grammar? If not, is there a way to use outlines to do this some other way? If it requires new features, what is the right way to explain the feature request?

[swamidass]

This is related to (but distinct) from a more common challenge of parsing HTML or XML with arbitrary tags.

These are valid:

<div></div>
<custom></custom>

But these are not:

<div></custom>
<custom></div>