fix: handle semantic flow edge cases

[lewis6991]

Problem:

• Several semantic-flow edge cases produced false diagnostics or imprecise types.
• The fixes cover union array keys, empty generic varargs, non-terminating return paths, guarded pcall callbacks, and call-condition replay.

Solution:

• Keep the fixes split into focused commits with targeted regressions.
• Tighten flow replay and branch narrowing where the existing model already had the right seam.

Testing:

• cargo test -p emmylua_code_analysis

[github-actions]

Code Review Summary

Changes Overview

This PR modifies the return type of analyze_func_body_missing_return_flags_with from (bool, bool, bool) to (bool, bool), removing the is_infinite flag. It also includes several test additions and semantic inference improvements.

Issues Found

1. Inconsistent Function Signature Across Files

• Problem: The function analyze_func_body_missing_return_flags_with is defined in three files (func_body.rs, mod.rs, compilation/mod.rs) with identical signatures. This violates DRY principle and creates maintenance burden.
• Suggestion: Consider consolidating to a single definition or using a re-export pattern.

2. Potential Logic Regression in check_return_count.rs

• Problem: The condition !can_fall_through && !can_break replaces the previous !can_fall_through && !can_break && !is_infinite. This may cause false positives for functions that are truly infinite but not detected by can_fall_through or can_break.
• Suggestion: Verify that the removal of is_infinite doesn't miss cases where a function has no return path but isn't captured by the remaining two flags.

3. Test Assertion Change in check_return_count_test.rs

• Problem: Line 309 changes assert!(!ws.has_no_diagnostic(...)) to assert!(ws.has_no_diagnostic(...)) for MissingReturn diagnostic. This is a significant behavioral change that should be documented.
• Suggestion: Add a comment explaining why this test expectation changed.

4. New Function infer_call_arg_type Logic

• Problem: The condition !cache.is_no_flow() || !matches!(&arg_expr, LuaExpr::TableExpr(_)) is complex and may be error-prone. The double negative makes it hard to understand.
• Suggestion: Simplify to: if cache.is_no_flow() && matches!(&arg_expr, LuaExpr::TableExpr(_)) { ... } else { infer_expr(...) }

5. Potential Performance Impact

• Problem: The new key_type_matches function in infer_array.rs calls check_type_compact which may be expensive for frequently accessed array indices.
• Suggestion: Consider caching results or adding a fast-path for common integer types before calling check_type_compact.

6. Missing Error Handling in infer_call_arg_type

• Problem: The function returns InferResult but doesn't handle the case where infer_expr returns an error after with_replay_overlay.
• Suggestion: Add proper error propagation or fallback.

Positive Aspects

• ✅ Comprehensive test coverage for new functionality
• ✅ Clear separation of concerns in the ExprTypeContinuation::Truthiness variant
• ✅ Good documentation of the MissingReturn logic change in comments

Recommendations

1. Add a changelog entry explaining the behavioral change in MissingReturn diagnostic
2. Consider adding integration tests for edge cases where is_infinite was previously required
3. Document the rationale for removing is_infinite from the return tuple

[gemini-code-assist]

Code Review

This pull request refactors the function body analysis by removing the is_infinite flag from return flag tracking, simplifying the logic for detecting missing return statements. It also improves type inference for pcall return values, array indexing with integer literal unions, and variadic generic parameters, specifically handling empty vararg sequences more accurately. Additionally, the PR enhances flow-sensitive type narrowing for call expressions and adds comprehensive test cases for these scenarios. I have no feedback to provide as there were no review comments.