Ship the Rust core with fat LTO and stripped symbols

[FarhanAliRaza]

What

Switches the release profile from thin to fat LTO and strips symbol tables from the shipped cdylib:

[profile.release]
lto = "fat"
codegen-units = 1
strip = true

Measured locally (Linux x86_64)

• Binary size: 1.51 MB -> 1.29 MB (-15%), which shrinks every wheel since stripping happens at link time.
• Runtime: no change on scripts/bench_scatter_native.py — 10M-point density prep reads 7.2–8.5 ms across repeats on both builds; run-to-run noise (±15%) dominates. CodSpeed on this PR is the cleaner verdict.
• Build time: 11.6 s -> 13.2 s.
• scripts/abi_smoke.py passes 140/140 against the stripped build — strip removes .symtab/debuginfo only, and .dynsym (what ctypes binds against) survives.

Deliberately not changed

• panic stays unwind: the C-ABI backstop in lib.rs converts kernel panics into sentinel returns via catch_unwind; panic = "abort" would abort the embedding CPython process. The wasm target keeps its explicit -C panic=abort in release.yml, where unwinding is unsupported anyway.
• No target-cpu bump: wheels stay baseline x86-64 (manylinux_2_17); AVX2 is already runtime-dispatched in src/simd.rs.
• No PGO yet: recorded in spec/design/rust-engine.md §2 as an open lever needing per-target training workloads in the release matrix.

Spec updated (spec/design/rust-engine.md §2) with the profile, the measurements, and the panic constraint; changelog entry under Unreleased.

Summary by CodeRabbit

• Performance

  • Reduced the shipped native library size by approximately 15%, from 1.51 MB to 1.29 MB.
  • Native scatter benchmark performance remains unchanged.

• Documentation

  • Added documentation for release build settings and platform-specific panic-handling behavior.

[coderabbitai]

📝 Walkthrough

Walkthrough

The Rust release profile now uses fat LTO, one codegen unit, and symbol stripping. Design documentation and the changelog record the panic-unwinding constraints and the resulting cdylib size reduction.

Changes

Rust release profile optimization

Layer / File(s)	Summary
Release profile settings and documentation Cargo.toml, spec/design/rust-engine.md, CHANGELOG.md	Release builds enable fat LTO, one codegen unit, and symbol stripping; documentation records FFI panic-unwinding requirements, binary-size results, and the changelog entry.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Suggested reviewers: alek99, masenf

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: enabling fat LTO and stripping symbols in the Rust core release build.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch release-profile-fat-lto

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@CHANGELOG.md`:
- Around line 73-76: Update the Rust core release-profile changelog entry to
document all three settings, including codegen-units = 1, and narrow the
stripping description to state that symbol tables are stripped while preserving
.dynsym. Match the terminology and rationale used in spec/design/rust-engine.md,
while retaining the size and benchmark results.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: da24bbcc-af73-4a2c-875e-1170755eafcf

📥 Commits

Reviewing files that changed from the base of the PR and between c1233a3 and 867c887.

📒 Files selected for processing (3)

• CHANGELOG.md
• Cargo.toml
• spec/design/rust-engine.md

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Document all release-profile changes and narrow the stripping claim.

The implementation changes three profile settings, but this entry omits codegen-units = 1. “Strips symbol tables” is also broader than the specification, which says .dynsym is preserved. Please align the wording with spec/design/rust-engine.md.

Proposed wording

-- The Rust core's release profile now uses fat LTO and strips symbol tables,
+- The Rust core's release profile now uses fat LTO, one codegen unit, and
+  strips `.symtab`/debug information while preserving dynamic exports,
   shrinking the shipped cdylib ~15% (1.51 → 1.29 MB) with no measured runtime

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	- The Rust core's release profile now uses fat LTO and strips symbol tables,
	shrinking the shipped cdylib ~15% (1.51 → 1.29 MB) with no measured runtime
	change on the native scatter bench (`spec/design/rust-engine.md` §2 records
	the profile and why `panic` stays unwinding).
	- The Rust core's release profile now uses fat LTO, one codegen unit, and
	strips `.symtab`/debug information while preserving dynamic exports,
	shrinking the shipped cdylib ~15% (1.51 → 1.29 MB) with no measured runtime
	change on the native scatter bench (`spec/design/rust-engine.md` §2 records
	the profile and why `panic` stays unwinding).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@CHANGELOG.md` around lines 73 - 76, Update the Rust core release-profile
changelog entry to document all three settings, including codegen-units = 1, and
narrow the stripping description to state that symbol tables are stripped while
preserving .dynsym. Match the terminology and rationale used in
spec/design/rust-engine.md, while retaining the size and benchmark results.

[codspeed-hq]

Merging this PR will not alter performance

✅ 103 untouched benchmarks
⏩ 2 skipped benchmarks¹

---

_{Comparing release-profile-fat-lto (867c887) with main (c1233a3)}

Footnotes

1. 2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩