perf(codegen): share numeric guards across dynamic add trees

[proggeramlug]

Summary

• recognize dynamic + trees with at least three leaves
• lower every leaf once, preserve the original tree associativity, and use one shared all-number guard for a native fadd arm
• preserve the original js_dynamic_string_or_number_add tree as the cold arm for strings, BigInt, Symbols, and object coercion
• keep two-leaf dynamic additions on their existing direct helper path, where a separate guard diamond does not pay for itself

Why

Accumulator expressions such as sum += row.x + row.y commonly reach codegen with all three leaves boxed as Any, especially across callback and interface boundaries. Lowering each + node independently then calls the full dynamic addition helper twice even when all runtime values are numbers.

The compiler already has a guarded numeric-tree lowering that evaluates leaves once and reconstructs the source tree in both arms. This change admits fully dynamic trees only when they have three or more leaves, which is the first point where one shared guard can replace multiple dynamic helpers.

This PR is stacked on #8674 (which is stacked on #8672). Please review/merge those first; this PR contains only commit ed9d86bce relative to its base, followed by the required changelog fragment.

Correctness

• cargo fmt --all -- --check
• focused IR tests prove:
  • a three-leaf dynamic tree has one numeric arm with two fadds and retains two dynamic helpers in the cold arm
  • a two-leaf dynamic add remains unchanged
• cargo test -p perry-codegen --lib: 1,191 passed, 0 failed, 1 ignored
• standalone Node/Perry fixture matched exactly for:
  • numeric trees
  • left- and right-associated string-sensitive results
  • BigInt trees and mixed-BigInt TypeErrors
  • Symbol TypeErrors
  • observable Symbol.toPrimitive ordering in both associativities
• the standalone fixture also passed with PERRY_GC_FORCE_EVACUATE=1 PERRY_GC_VERIFY_EVACUATION=1
• unchanged full codehz/ecs comprehensive suite: 7/7 passed with exact 50,005,000 accumulation checksum
• retained AArch64 assembly has two fadds on the numeric arm and both original dynamic-helper calls on the cold arm

The exact committed compiler emitted the same __text SHA-256 as the benchmarked artifact (b19ad70a31fa27d93f26e50bf1d140e5d577ebc9a0c256b3a9d762855f9fc90a).

Performance

Apple M1 Mac mini on AC power. Each cohort followed a 60-sample quiet-host gate with every sample at or below 25% active CPU, used taskpolicy -t 0 -l 0, and alternated process order.

Against the exact #8674 code (repeat=64), 11 accumulation pairs measured:

codehz/ecs kernel	#8674 median	candidate median	median paired reduction	wins
10k accumulation query	0.26989 ms/op	0.25891 ms/op	4.11%	11/11

All 22 processes passed the 7/7 suite oracle and exact checksum. A separate three-pair read-only regression check was neutral at the paired median (-0.033%, 1/3 wins); this optimization does not target that callback.

This is a measured incremental improvement, not a claim of Node parity. The remaining gap is dominated by the guarded ECS loop body and indirect boxed callback boundary.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b0d086c5-72c2-4746-9425-e107294725fe

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[proggeramlug]

Landed on main via #8682 (squash 67f3f7501).

Your branch was stacked on #8672, but the contents touch disjoint files, so this commit was cherry-picked onto main on its own — #8672 itself is held pending a rebase (details on that PR).

Validated on the merged result: 9 ratchet gates + cargo fmt --check pass, perry-codegen --lib 1198/0, perry-runtime --lib 2652/0 at RUST_TEST_THREADS=1.

One adjustment: #8674's new test pushed array/tests.rs from 1998 to 2034 lines, past the 2000-line gate, so the growth-forwarding group was relocated into an array/forwarding_tests.rs sibling (matching the existing subclass_tests pattern). Pure relocation — all 5 moved tests verified still running. Thanks!