ruff_cpp_spo: harvest_network example — sink Tesseract LSTM network C++ to SPO

[AdaWorldAPI]

Summary

The ruff→OGAR sink-in for the Tesseract recognizer transcode. A new ruff_cpp_spo example, harvest_network, walks Tesseract's 11 LSTM network-layer headers (network.h + the FullyConnected/LSTM/Series/Parallel/Plumbing/Convolve/Maxpool/Reversed/Reconfig/Input subclasses) via libclang and emits the has_function / inherits_from / virtually_overrides SPO manifest — the classid → ClassView method-resolution table that the downstream network→V3-SoA sink (lance-graph-contract::network) and the invoke_network dispatch resolve against. It mirrors the proven harvest_unicharset example (same walk_tu → model_from_class → expand → to_ndjson pipeline; corpus stays UPSTREAM, never vendored).

The manifest is the "vtable the hand-rolled enum would have faked": the Forward override set (FullyConnected/LSTM/Series/Parallel/Convolve/Maxpool/Reversed/Reconfig/Input) is the recognizer's compute-leaf list; the DeSerialize override set (FullyConnected/LSTM/Plumbing/Convolve/Maxpool/Reconfig/Input) is the binary-leaf list.

Additive: one new [[example]] (gated required-features = ["libclang"], so a default --all-targets build never compiles it) + the example file. No library change.

Test Plan

Ran against real Tesseract 5.5.0 source:

TESSERACT_SRC=/tmp/tesseract LIBCLANG_PATH=/usr/lib/llvm-18/lib \
  cargo run -p ruff_cpp_spo --features libclang --example harvest_network
# → 62 unique classes across 11 headers; 5060 triples → /tmp/network_manifest.ndjson
# → per-class Forward/DeSerialize override manifest printed for all 11 layer types

cargo clippy -p ruff_cpp_spo --features libclang --all-targets -- -D warnings clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_016b33swuXE23hKtqxsHu9p1

---

Generated by Claude Code

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a3cc828. Configure here.}

[cursor]

Libclang failures exit successfully

Medium Severity

When walk_tu returns WalkError::Libclang (missing LIBCLANG_PATH, libclang unavailable), the example logs to stderr and continues instead of failing. It can still write an empty or partial manifest and return Ok(()), unlike harvest_unicharset and the crate’s walk_files helper, which propagate that error.

 

^{Reviewed by Cursor Bugbot for commit a3cc828. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a3cc82880b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Fail when libclang harvesting fails

When LIBCLANG_PATH is wrong or libclang cannot be loaded, every walk_tu call returns WalkError::Libclang; this arm only logs the error and continues, so the command exits successfully and writes an empty or partial manifest for downstream consumers. Treat non-recoverable walker errors as fatal (as the library extraction path does) so a bad environment cannot silently produce a bogus network manifest.

Useful? React with 👍 / 👎.