fix: config reader/writer, formatter, and ExtrasError bugs

[henryiii]

🤖 AI text below 🤖

Part of #1357

Fixes a batch of verified Config/Formatter/Error bugs. Each was reproduced before changing; regression tests accompany the behavioral fixes.

include/CLI/impl/Config_inl.hpp

1. Section header with a trailing comment is now recognized. [section] # a comment (legal TOML) previously fell through the front()=='[' && back()==']' check (because back() was 't', not ']'), producing a bogus flag item named [section] and mis-attributing following keys. The reader now strips a quote-aware trailing comment for lines that open with [ but do not end with ], before the section check. Value lines are untouched (they still handle their own trailing comments downstream).
2. A one-line multiline comment no longer swallows the rest of the file. '''comment on one line''' followed by real config now parses correctly. After detecting the '''/""" opener the same line is checked for a closer, guarded by a length >= 6 requirement so the opening quotes themselves are not counted as the closer.
3. Off-by-two in Join-policy embedded-delimiter detection. A joined string of N elements has N-1+e delimiters; the array fallback should trigger when e > 0, i.e. opt->count() < delim_count + 1. The code tested < delim_count - 1, so embedded-delimiter cases slipped through and wrote a bare join (e.g. vals="c,[a,b]") instead of the array form (vals=['c', "[a,b]"]). Fixed the comparison (signedness preserved via the existing cast). A round-trip test exercises the array-fallback path via the [[ ]] vector escape, which is the reachable way an element retains an embedded delimiter (normal delimiter-splitting options never produce e > 0).
4. clean_name_string namespace/signature mismatch. Config.hpp declared void detail::clean_name_string(std::string&, const std::string&), but the definition lived in the public CLI:: namespace and returned std::string&, so the detail declaration was never defined and the real function polluted CLI::. The definition is moved into namespace detail, the return type reconciled to void (matching the declaration; no caller used the return value), and the four call sites in to_config qualified as detail::clean_name_string.
5. Consistent comment lead on group headers. The unnamed-subcommand group header wrote only commentLead (# Group Options); it now matches the main option-group header style commentChar << commentLead (## Group Options).

include/CLI/impl/Formatter_inl.hpp

6. Hidden positionals no longer render as [] in the usage line. make_usage now filters positionals with !opt->get_group().empty(), matching make_positionals.
7. Long subcommand names no longer glue onto the description. make_subcommand now mirrors make_option's long-name branch: when the name fills the column it emits a newline and disables the skip-first-line-prefix, so the description starts on its own indented line.
8. Non-paragraph description path no longer emits an extra blank line. make_description already returns desc + "\n\n"; the enable_description_formatting(false) branch in make_help dropped its extra '\n' so toggling the flag no longer changes spacing. (Scope limited to make_help per the report; make_expanded left as-is.)

include/CLI/Error.hpp

9. ExtrasError(name, args) reports the right error name. The two-argument constructor previously delegated to the protected 3-arg form, making get_name() return the app/subcommand name instead of "ExtrasError" (every other error class returns its class name, and App::exit/user code dispatch on get_name()). It now folds the app name into the message and constructs via the public single-arg path so get_name() == "ExtrasError", while preserving (and slightly enriching) the message content.

Tests

• tests/ConfigFileTest.cpp: TomlSectionTrailingComment (1), TomlDocStringCommentOneLine (2), IniJoinEmbeddedDelimiterRoundTrip (3, verified to fail on the pre-fix comparison).
• tests/FormatterTest.cpp: HiddenPositionalUsage (6), LongSubcommandName (7), DescriptionNoFormattingSpacing (8) — all verified to fail without their fix.
• tests/AppTest.cpp: ExtrasErrorName (9).
• Items 4 and 5 are covered by existing config round-trip tests, which continue to pass.

No false positives were found; all nine issues reproduced.

Verification

ctest over the full suite: 24/24 passed. The five primary binaries (ConfigFileTest, HelpTest, FormatterTest, AppTest, SubcommandTest) all pass; no existing tests required changes. prek -a --quiet is clean.

🤖 Generated with Claude Code