Add inline repetition to compound specs

[kendonB]

Summary

• add inline repetition support for compound specs using open-ended brace ranges
• propagate repeated extras through Compound, CompoundRule, MappingRule, and backend compilers
• add parser, compiler, rule, and doctest coverage for the new repetition behavior

Testing

• .venv/bin/python -m pytest dragonfly/test/test_lark_parser.py dragonfly/test/test_compound_rule_repetition.py dragonfly/test/test_compiler_natlink.py dragonfly/test/test_compiler_kaldi.py dragonfly/test/test_compiler_sapi5.py dragonfly/test/test_compiler_sphinx.py
• note: the legacy natlink binary snapshot assertion in test_compiler_natlink.py still mismatches locally

---

Note

Medium Risk
Touches core parsing and repetition decoding plus multiple engine compilers, so regressions could affect recognition matching/greediness or engine-specific grammar generation; changes are well-covered by new targeted tests.

Overview
Adds inline repetition syntax to compound specs via brace ranges (e.g. "<item>{1,}", "{0,}", "{2,4}"), with parser updates to distinguish repetition quantifiers from existing {weight=...}-style specifiers and to avoid mutating shared extras when applying specials.

Extends Repetition with an unbounded mode (validated against empty-matchable children), updates its runtime decode()/get_repetitions() behavior, and propagates repeated extras as lists through Compound value functions and through CompoundRule/MappingRule via new shared extract_compound_extras() logic (including default/omission handling and per-branch shape inference).

Updates Kaldi/Natlink/SAPI5/Sphinx compilers to emit correct grammar structures for unbounded repetition (preserving min and optional skip behavior) and adds substantial unit/doctest coverage for the new parsing, repetition, and compiler semantics.

^{Written by Cursor Bugbot for commit 944e0b6. This will update automatically on new commits. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 944e0b6afd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Handle unbounded Repetition in element_matches_empty

element_matches_empty() currently falls through to the generic Sequence branch for Repetition, which makes Repetition(..., min=0, unbounded=True) look non-empty whenever its child is non-empty. That lets nested grammars like Repetition(Repetition(Literal("a"), min=0, unbounded=True), min=0, unbounded=True) bypass the constructor guard, even though the inner repetition can match zero words; at runtime this can create non-terminating decode loops because the outer unbounded repetition can keep accepting zero-length child matches.

Useful? React with 👍 / 👎.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Empty-check misses zero-min Repetition

High Severity

element_matches_empty treats Repetition as a plain Sequence, so unbounded repetitions with min=0 are considered non-empty. This bypasses the constructor guard and allows unbounded repetition children that can accept empty, causing non-consuming repetition loops.

Additional Locations (2)

• dragonfly/grammar/elements_basic.py#L663-L665
• dragonfly/grammar/elements_basic.py#L753-L757