Fix TS1515 not reported when duplicate named group is nested inside a group

[Copilot]

scanDisjunction dropped an alternative's named-capture scope as soon as the alternative (or enclosing group) ended, so a duplicate (?<name>...) defined inside a nested group never reached the enclosing alternative's duplicate check.

/(?<a>x)(?<a>y)/       // TS1515 (already worked)
/(?:(?<a>x))(?<a>z)/   // no error before this fix — should be TS1515
/(?=(?<a>x))(?<a>z)/   // no error before this fix — should be TS1515
/(?<a>x)|(?<a>y)/      // legal, must stay quiet

Root cause

• scanDisjunction pushed a fresh scope per alternative (including alternatives inside nested groups/assertions), but discarded that scope entirely once the alternative or group closed, instead of folding it back into the enclosing alternative.

Fix

• scanDisjunction (internal/scanner/regexp.go) now unions the names defined across all alternatives of a disjunction, and when that disjunction is the content of a nested group, merges the union up into the parent alternative's scope before returning.
• This makes names defined inside non-capturing groups, capturing groups, and lookahead/lookbehind assertions visible to later sibling terms in the same alternative, while leaving names confined to separate branches of a | disjunction untouched (still legal per ES2025).

Tests

• Added testdata/tests/cases/compiler/regexNamedGroupDuplicateNestedInGroup.ts covering all patterns from the issue, with accepted baselines showing TS1515 now fires for the previously-missed nested cases and remains silent for the legal alternation cases.

• Fixes TS1515 is not reported when the earlier duplicate named group is inside a nested group #4878

[Ryan Cavanaugh (RyanCavanaugh)]

Paging graphemecluster (@graphemecluster), this is above my regex paygrade

[graphemecluster (graphemecluster)]

I checked my old fix from microsoft/TypeScript@c435526 (#60249) _{(yes sorry I should've extracted the change and fired a separate PR in the past)} and it implements the same logic (unioning the nested alternatives into the parent when the disjunction closes), so this should be fine.

I also verified using the test case file I made up in the past (microsoft/TypeScript@24c16f9 (#60249)) and it passes. The current PR's test coverage is quite narrow (e.g. it does not test reverse direction), so I would suggest adding my cases in that commit, or at least:

// Named group in non-first alternative
/(?:x|(?<a>y))(?<a>z)/;

// Outer named group before nested named group
/(?<a>x)(?:(?<a>y)|z)/;

// Named group inside modifier group
/(?i:(?<a>x))(?<a>y)/;

// Multiple nesting levels
/(?:(?:(?<a>x)))(?<a>y)/;

// Legal: Outer disjunction separates non-nested and nested named groups
/(?:(?<a>x))|(?<a>y)/;

// Legal: Outer disjunction separates each duplicate pair
/(?<a>x)(?<b>y)|(?<a>z)(?<b>w)/;

On the other side, (?<\u0061>...) currently fails with TS1514 but not normalised to a and participated in the duplicate check. I am porting microsoft/TypeScript#61042 to fix this.

[Copilot]

Pull request overview

Fixes nested named-capture scope propagation so TS1515 is correctly reported without affecting legal alternations.

Changes:

• Unions nested disjunction capture names into the enclosing alternative.
• Adds regression coverage for nested groups and assertions.
• Adds generated compiler baselines confirming five diagnostics.

Show a summary per file

File	Description
internal/scanner/regexp.go	Propagates nested capture names.
testdata/tests/cases/compiler/regexNamedGroupDuplicateNestedInGroup.ts	Adds regression cases.
testdata/baselines/reference/compiler/regexNamedGroupDuplicateNestedInGroup.errors.txt	Records expected diagnostics.
testdata/baselines/reference/compiler/regexNamedGroupDuplicateNestedInGroup.js	Records emitted JavaScript.
testdata/baselines/reference/compiler/regexNamedGroupDuplicateNestedInGroup.symbols	Records symbol baseline.
testdata/baselines/reference/compiler/regexNamedGroupDuplicateNestedInGroup.types	Records type baseline.

Review details

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 6/6 changed files
• Comments generated: 0
• Review effort level: Balanced

[Jake Bailey (jakebailey)]

This really should have been var disjunctionNames collections.Set[string] but it's whatever I geuss