Implement Kotzmann VEE'05 connection graph escape analysis for arrays

[Copilot]

Implements intraprocedural escape analysis following Kotzmann & Mössenböck's VEE'05 approach for array allocations in ZJIT HIR.

Changes

Three-level escape classification

• NoEscape: Local-only (scalar replacement candidate)
• ArgEscape: Returned/passed to callees (limited optimization)
• GlobalEscape: Heap escape via ivars/globals (no optimization)

Connection graph implementation

• Nodes track allocation sites with escape states
• Points-to edges model array element relationships (ArrayAref/ArrayAset)
• Phantom nodes handle unknown references conservatively
• Fixed-point propagation (typically 2-5 iterations, max 100)

Analysis phases

1. Build connection graph from array allocations
2. Add edges: method calls→ArgEscape, ivar stores→GlobalEscape, returns→ArgEscape
3. Propagate states through points-to relationships
4. Extract final classifications

Integration with existing infrastructure

• Leverages LocalArray effect for DCE of NoEscape arrays
• Compatible with PatchPoint deoptimization
• Returns HashMap<InsnId, EscapeState> for future optimizations

Example

// Array never leaves method - NoEscape
def local_only
  arr = [1, 2, 3]
  arr.length
end

// Array returned to caller - ArgEscape  
def returns_array
  [1, 2, 3]
end

// Array stored to heap - GlobalEscape
def stores_array
  @arr = [1, 2, 3]
end

Paper: https://www.usenix.org/legacy/events/vee05/full_papers/p111-kotzmann.pdf

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.