Combined: pre-allocate source shapes + singleton proxy handler + cache proxyOf in each()

[alanzabihi]

Hypothesis

Three near-miss optimizations showed directional improvement individually:

• Pre-allocate Node source object structure to reduce shape transitions #53: pre-allocate source object shape: -2.61ms
• Avoid repeated proxyOf and parent dereferences in container.each hot loop #52: singleton proxy handler + cache proxyOf/indexes in each(): -3.14ms (high variance)
• Inline container.each into walkSync and eliminate iterator allocation #44: inline each() in walkSync: -1.91ms

The deltas dont add linearly, but combining the two strongest (#52 and #53) both target object shape/allocation overhead. Together they reduce V8 hidden class transitions AND repeated property resolution.

Implementation:

1. Pre-allocate source objects with {start, end: null, input} in parser (from Pre-allocate Node source object structure to reduce shape transitions #53)
2. Singleton proxy handler at module level in container.js (from Avoid repeated proxyOf and parent dereferences in container.each hot loop #52)
3. Cache proxyOf and indexes references in each() (from Avoid repeated proxyOf and parent dereferences in container.each hot loop #52)

These are independent optimizations — no interference expected. Combined delta target: 4-5ms.

Editable surface

• lib/parser.js — pre-allocate source object shape
• lib/container.js — singleton proxy handler, cache proxyOf/indexes in each()
• lib/node.js — singleton proxy handler for leaf nodes

What's different from prior work

• Pre-allocate Node source object structure to reduce shape transitions #53 alone: -2.61ms (below 3ms). Avoid repeated proxyOf and parent dereferences in container.each hot loop #52 alone: -3.14ms (high variance). Combining eliminates the variance concern by providing a larger signal. Neither overlaps in code paths.
• Combined: inline each() in walkSync + pre-compute keyed visitor maps to compound sub-tolerance gains #56 combined walkSync inlining with keyed maps — different optimization targets.

Expected impact

Combined METRIC composite improvement of 4-6ms. The two strongest near-misses target complementary overhead (allocation shapes vs property resolution).

[alanzabihi]

Polyresearch approval: thesis #66.