perf(cli): mimalloc global allocator + build ariadne SourceCache once per diagnostic batch

[hellovai]

Re-lands the baml_cli slice of the cold-compile performance audit (#4016, commit c1466f3), re-derived against current canary. Two independent changes:

1. mimalloc as baml_cli's global allocator. The compiler workload is dominated by small short-lived allocations (Ty trees, Vecs, SmolStrs); system malloc measured ~35% of remaining single-threaded CPU in the original audit.
2. Build the ariadne SourceCache once per diagnostic batch in render_diagnostics instead of once per diagnostic (each Source::from computes a line index over every project file), threading it as &mut through render_ariadne / render_report_to_string. Rendered output bytes are unchanged.

Measurements

CLI wall time on the warning-heavy corpus (baml check crates/baml_tests/baml_src, ~100 warnings), disk cache disabled via BAML_NO_BYTECODE_CACHE=1 + fresh BAML_CACHE_DIR, release builds of clean canary (2660b8b) vs. this branch, alternated in one hyperfine invocation:

Benchmark 1: BEFORE
  Time (mean ± σ):      3.947 s ±  0.587 s    [User: 3.118 s, System: 0.079 s]
  Range (min … max):    3.080 s …  5.032 s    10 runs

Benchmark 2: AFTER
  Time (mean ± σ):      1.304 s ±  0.280 s    [User: 1.182 s, System: 0.037 s]
  Range (min … max):    1.119 s …  2.012 s    10 runs

Summary
  AFTER ran
    3.02 ± 0.79 times faster than BEFORE

The benchmark machine was under heavy background load, hence the wide σ; user CPU time (3.12 s → 1.18 s, ~2.6x) is load-independent and consistent across all runs.

Cache-verify gate (#3924 compatibility)

Cached diagnostics are replayed byte-for-byte on cache hits, so rendered output must not change. Verified with BAML_CACHE_VERIFY=1 and a fresh BAML_CACHE_DIR (cache enabled), two runs each on before/after binaries so the second run replays cached diagnostics:

• no divergence reported, all runs exit 0;
• captured stdout+stderr of before vs. after binaries is byte-identical (only the wall-time suffix of the final "Finished checked 77 file(s) in Ns" status line differs between runs);
• run 1 (cold) vs. run 2 (cached replay) output identical for both binaries.

Tests

• cargo test --workspace: passes. (One caveat: three targets initially failed for environmental reasons on the shared bench machine — pack_e2e hit ENOSPC while the disk was full, the Python SDK cancellation tests need Python ≥ 3.11 (ExceptionGroup, asyncio.timeout) and uv had picked 3.10, and the TS fixtures were missing node_modules after the disk filled up. All three pass after freeing disk / UV_PYTHON=3.12 / re-running setup.sh, with no relation to this change.)
• Targeted re-run on the rebased tip: baml_cli (385 tests), baml_compiler_diagnostics (18), baml_project — all pass.
• Pre-commit hooks (cargo fmt, cargo stow, workspace clippy -D warnings) pass.
• No snapshot changes.

Notes

• Expected trivial merge conflict with chore(tools): add tools_compile_profile — salsa/pipeline profiling harness #4038 (in-flight profiling tool), which adds the identical mimalloc = { version = "0.1.52" } line to [workspace.dependencies].
• Provenance: perf(compiler2): ~32x faster cold compile — remove redundant work, memoize package-invariant queries #4016 (original audit branch perf/compiler2-cold-compile).

Summary by CodeRabbit

• Performance
  • Improved cold-start performance for baml check and baml build.
  • Improved rendering performance when displaying multiple compiler diagnostics.
  • Reduced repeated processing while generating Ariadne-formatted diagnostic output.
• Bug Fixes
  • Preserved existing fallback behavior when a diagnostic cannot be rendered.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
beps	Ready	Preview, Comment	Jul 16, 2026 2:19am
promptfiddle	Ready	Preview, Comment	Jul 16, 2026 2:19am
promptfiddle2	Ready	Preview, Comment	Jul 16, 2026 2:19am

[github-actions]

⏭️ Performance benchmarks were skipped

Perf benchmarks (CodSpeed) are opt-in on pull requests — they no longer run on every push. They always run automatically after merge to canary/main.

To run them on this PR, do any of the following, then push a commit (or re-run CI):

• Add RUN_CODSPEED=1 to the PR description, or
• Include run-perf or /perf in the PR title or any commit message.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 3b3dc087-c05e-48e5-b761-e23beee0298e

📥 Commits

Reviewing files that changed from the base of the PR and between 11e75dc and 32e0507.

⛔ Files ignored due to path filters (1)

• baml_language/Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (4)

• baml_language/Cargo.toml
• baml_language/crates/baml_cli/Cargo.toml
• baml_language/crates/baml_cli/src/main.rs
• baml_language/crates/baml_compiler_diagnostics/src/render.rs

---

📝 Walkthrough

Walkthrough

The CLI now uses mimalloc as its global allocator. Ariadne diagnostic rendering creates a source cache once and reuses it across single or batch rendering paths.

Changes

Performance updates

Layer / File(s)	Summary
CLI allocator integration baml_language/Cargo.toml, baml_language/crates/baml_cli/Cargo.toml, baml_language/crates/baml_cli/src/main.rs	Adds the pinned mimalloc workspace dependency, enables it in baml_cli, and configures it as the global allocator.
Shared Ariadne source cache baml_language/crates/baml_compiler_diagnostics/src/render.rs	Creates and reuses SourceCache across Ariadne diagnostic rendering while preserving concise-format rendering and existing error fallback behavior.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• BoundaryML/baml#4016: Updates the same CLI allocator and Ariadne source-cache rendering logic.

Suggested reviewers: antoniosarosi, sxlijin

Poem

I’m a rabbit with a cache to share,
And mimalloc bouncing through the air.
Diagnostics stream with sources bright,
The CLI compiles cold and light.
Hop, hop—fewer builds anew!

🚥 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 two main performance changes: mimalloc in the CLI and shared Ariadne SourceCache batching.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch perf/reland-cli-alloc-diag-render

---

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.}

[github-actions]

Binary size checks passed

✅ 7 passed

	Artifact	Platform	File	Gzip	Gated on	Baseline	Delta	Status
✅	baml-cli	Linux	🔒 23.0 MB	9.8 MB	file	22.8 MB	+212.4 KB (+0.9%)	OK
✅	packed-program	Linux	🔒 16.3 MB	6.8 MB	file	16.3 MB	+24.6 KB (+0.2%)	OK
✅	baml-cli	macOS	🔒 17.7 MB	8.5 MB	file	17.5 MB	+185.9 KB (+1.1%)	OK
✅	packed-program	macOS	🔒 12.6 MB	6.0 MB	file	12.6 MB	+32 B (+0.0%)	OK
✅	baml-cli	Windows	🔒 19.1 MB	8.7 MB	file	18.7 MB	+398.8 KB (+2.1%)	OK
✅	packed-program	Windows	🔒 13.5 MB	6.1 MB	file	13.5 MB	+5.6 KB (+0.0%)	OK
✅	bridge_wasm	WASM	15.2 MB	🔒 4.3 MB	gzip	4.3 MB	+3.6 KB (+0.1%)	OK

🔒 = the size this artifact is GATED on (ceiling + delta). Binaries gate on file size (installed binary); WASM gates on gzip (download size). The other size is shown for information only.

---

Generated by cargo size-gate · workflow run

[hellovai]

Folded into the combined re-landing PR #4054 (per maintainer preference for a single PR post-tool-merge). Branch kept for provenance; individual before/after measurements remain in this PR's description.