FIX Preserve word-selection parameters in StringJoinConverter identifiers

[Joel (aspire488)]

Description

Tests and Documentation

Description

Fixes #2973.

"StringJoinConverter._build_identifier()" previously replaced the identifier parameters from "WordLevelConverter" instead of preserving them. This caused converters with different "word_selection_strategy" configurations to produce the same identifier and registry name, even though they can produce different outputs.

Changes

• Preserve the parent "WordLevelConverter" identifier parameters when building the "StringJoinConverter" identifier.
• Keep "join_value" as the converter-specific parameter.
• Add regression coverage for:
  • distinct word-selection configurations producing distinct identifiers
  • equivalent index selections producing the same identifier
  • independently registering converters with different selection configurations

Testing

Added focused unit tests in "tests/unit/converter/test_string_join_converter.py".

The PyRIT test suite was not executed in the connector environment; CI should provide full validation.

[Zhibo Lin (LE0-Lin)]

Thanks for working on this. I had a local fix and regression tests prepared when filing #2973 and was waiting for issue curation. Could we coordinate on the remaining coverage and the default-identifier compatibility question? The local work used Codex assistance.

[Joel (aspire488)]

Thanks for raising the compatibility concern. I’ve updated the PR to preserve the legacy identifier for the default StringJoinConverter configuration while still including word-selection parameters for non-default strategies.
I also added regression coverage for:
default identifier compatibility
explicit AllWordsSelectionStrategy compatibility
distinct selection configurations producing distinct identifiers
equivalent index sets producing the same identifier
registering distinct configurations independently

[Richard Lundeen (richlundeen)]

This fix looks good! Zhibo Lin (@LE0-Lin), also thank you for opening a great issue, which imo is just as important. I'll make sure to credit you in this merge and we can get it in

[Richard Lundeen (richlundeen)]

Reviewed with GitHub Copilot. This addresses #2973: distinct word selections have distinct identifiers and registry names, equivalent index sets keep the same identity, and the legacy default identity is preserved. Added coverage for default/partial registration, exact parameter contents, and custom all-words subclasses. All 15 scoped tests and all pre-commit hooks passed locally. Credit to Joel (@aspire488) for the fix and Zhibo Lin (@LE0-Lin) for the original diagnosis and reproduction; LE0-Lin is also credited as a co-author in the new commit. CI must pass before this enters the merge queue.

[Zhibo Lin (LE0-Lin)]

Thanks for the review and for crediting the original diagnosis and reproduction. I appreciate the coordination and the added compatibility coverage.