Detect line-level local-variable duplication inside spliced function bodies

[TheValiantOne]

Summary

The last compile error standing after the v0.6.3 pristine live merge: Variable 'mCSMCR' is already defined in merged combat.ws — DiffPlex's silent-duplication mode striking at line granularity, duplicating a mod's comment-wrapped local declaration block inside a function body. Every unit-level invariant from #33 passes such output (unit counts are all correct); the game's compiler doesn't.

Fix

• HasDuplicatedLocalVarDecls: counts local var declarations per name inside a unit's text (comment-stripped so commented-out declarations don't skew; multi-declarator lines count each name). The merged unit may never declare a local more often than every input version of the same unit — duplication an input already carried is tolerated (same only-flag-what-the-merge-introduced philosophy as the unit-count invariants).
• Wired into both splice sites: ValidateWholeFileMergeOutput (per matched scoped unit — a corrupt whole-file "clean" step now routes to the function-level rescue) and ResolveUnit's per-function 3-way path (a corrupt per-function splice falls back to the whole-function tiebreak, with an audit note naming the duplicated local).

Validation

3 new tests: the exact combat.ws shape detected; pre-existing input duplicates tolerated; commented-out declarations ignored. 156 total, green; format clean. Live re-merge of the affected file with this build posted below once run.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GXAuGMLB44T5Zv5o5ZzKah

[TheValiantOne]

Live repair confirmed

Re-merged the affected combat.ws on the real install with this branch's build. Two-step finding worth recording: the first overwrite re-merge kept the duplicate — correctly, because the corrupted merged file re-enters the chain as the accumulated source, making the duplicate a pre-existing input duplicate the invariant deliberately tolerates. The actual repair required a pristine chain for that file (drop its inventory record + delete the corrupted output, then merge vanilla+mods) — after which: exactly one var mCSMCR declaration, all usages intact, structural gate passes, output self-consistent under the full validator, 0 decisions needed (clean per-function resolution).

Practical takeaway for users hitting merge-corrupted output from older builds: refreshing with overwrite faithfully preserves the corruption (it's in the input); the fix is removing that file's merged output + inventory record and re-merging fresh. Worth a docs note when this merges.

🤖 Generated with Claude Code