tooling: per-alloc-site memory tracker in compare_ffi_memory.rs

[polaz]

Context

compare_ffi_memory.rs reports total bytes allocated by Rust vs FFI per scenario but doesn't attribute allocations to call sites.

Resolution

Implemented as the dhat-heap cargo feature (default off) wired into the standalone profiling examples instead of the criterion harness:

• zstd/Cargo.toml: optional dhat = "0.3" dep behind dhat-heap feature
• zstd/examples/encode_loop_reuse_z000033.rs, zstd/examples/encode_loop_dict.rs: global allocator + profiler blocks; write dhat-heap.json with per-call-site blocks/bytes (totals, t-gmax, t-end)

The standalone-example shape beats the original compare_ffi_memory.rs plan: the criterion harness pollutes attribution with setup/training/validation allocations, while the examples isolate the steady-state loop. Allocation COUNTS are platform-independent, so runs on any host diagnose musl-amplified churn.

Proven in practice: attributed the 13 allocs/frame on the reused-compressor dict path (row-table oscillation / estimator boxes / huffman seed clone / header temporaries) and the BT dms per-frame rebuild — both eliminated (musl 5.5× win).

The donor-proxy delta column (vs ZSTD_sizeof_CCtx) was not built; compare_ffi_memory.rs totals already give the cross-side comparison, dhat gives the our-side breakdown.

[coderabbitai]

🔗 Related PRs

#143 - bench: add real peak-alloc metric for full encode/decode/memory baseline [merged]
#282 - perf(decode): lazy ring-buffer allocation for direct-eligible frames [merged]
#301 - bench(dict): report Rust dict-compressed size + emit compress-dict ratio [merged]
#390 - perf(encode): reshape dfast match-find toward donor structure [merged]
#395 - perf(encode): zero per-frame allocations on the reused-compressor path [closed]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!