Code graph and memory for your coding agents, in one Rust program. For Claude Code, Codex, Cursor and any MCP client.
Upstream tests the parity checks do not run
The comparison checks of the graph engine run most of the test files
(tests/test_*.py) of Graphify 0.9.88 (commit a9860f70) against the engine.
A file counts as run when at least one of its tests runs and passes in a check,
or when the engine fails one of its tests and the check's receipt records that
as a known difference. This page lists the others, with the reason each one
is not run, so that the parity claim in
benchmarks.md can be read against
it. The known differences are in
upstream-tests-known-differences.md, and
every file's status is in
upstream-tests-0.9.88.json.
- Not applicable. The file tests something the engine does not contain, or something that exists only in Python or in Graphify's own test setup.
- Not yet run. The file tests behaviour the engine ports, and no check runs it yet. Other checks and the engine's own tests may cover some of that behaviour; this page does not claim that they do.
Against Graphify 0.9.77, from which the port was made, most of these files are unchanged; the tables mark the new and changed ones. "Covers" names the Graphify modules a file imports. The list was made on 14 October 2026 from the run of the widened checks on that day, which recorded every upstream test's outcome, and from its receipts.
Not applicable
| Test file | Since 0.9.77 | Why it is not run |
|---|---|---|
tests/test_atomic_canvas_export.py |
unchanged | It observes atomic writes by patching Python's os.replace, which the engine's writes do not call |
tests/test_backend_env_isolation.py |
unchanged | Graphify's own test fixture that clears the backend variables (it reads the Python source of detect_backend) |
tests/test_backend_extras.py |
changed | Python packaging: the pip extra of the claude backend and its install hint |
tests/test_cloud_cta.py |
new | Graphify Cloud's call to action, which the port leaves out |
tests/test_extractors_registry.py |
unchanged | Python module layout: graphify.extract re-exports the same function objects |
tests/test_home_sandbox.py |
unchanged | Graphify's own test fixture that redirects HOME |
tests/test_js_walk_order.py |
new | A speed-up inside a private Python function (the reference's _walk_js_tree cursor walk, not ported); no extraction runs |
tests/test_language_resolvers.py |
unchanged | It registers Python functions as resolvers; the engine's registry runs native passes, which check_resolver_registry drives with equivalent recorded scenarios |
tests/test_manifest_tomli_required.py |
unchanged | Python before 3.11 without tomli |
tests/test_multigraph_compat.py |
unchanged | It probes the Python runtime and NetworkX; check_multigraph compares the engine's own graph backend probe |
tests/test_pdf_extra_warning.py |
unchanged | It hides the optional pypdf package to test the missing-extra warning; the engine's PDF reader is built in, so the warning cannot occur |
tests/test_pin_hash_seed.py |
unchanged | Python's PYTHONHASHSEED, which the interpreter reads when it starts |
tests/test_python_parse_memoization.py |
unchanged | A speed-up inside private Python functions |
tests/test_serve_http.py |
changed | It skips unless the mcp and starlette packages are installed, and it drives the reference's HTTP app inside Python; check_native_workflows drives the engine's HTTP transport |
tests/test_skill_codex_semantic_merge.py |
new | The text of Graphify's own skill; the engine installs its own skill files |
tests/test_skill_semantic_all_cached.py |
unchanged | The text of Graphify's own skill; the engine installs its own skill files |
tests/test_skillgen.py |
changed | Graphify's skill generator (tools/skillgen) and the skill files it writes; the engine ships its own |
tests/test_skillgen_input_path_injection.py |
changed | Graphify's skill generator (tools/skillgen) and the skill files it writes; the engine ships its own |
tests/test_walk_python_tree_iterative.py |
unchanged | A speed-up inside private Python functions |
tests/test_wheel_packaging.py |
unchanged | Python packaging: it builds the wheel |
Not yet run
| Test file | Since 0.9.77 | Covers | Note |
|---|---|---|---|
tests/test_anthropic_custom_endpoint.py |
unchanged | llm |
|
tests/test_antigravity_install.py |
unchanged | command line | |
tests/test_atomic_version_stamp.py |
unchanged | install |
|
tests/test_claude_md.py |
unchanged | command line | |
tests/test_cli_broken_pipe.py |
unchanged | command line, as a subprocess | Starts the command line with subprocess.Popen, which the harness does not route to the engine |
tests/test_falkordb_integration.py |
unchanged | build, export |
|
tests/test_file_truncation_warning.py |
unchanged | llm |
|
tests/test_graphdb_push_indexes.py |
unchanged | exporters.graphdb |
|
tests/test_hook_chain_survives_skip.py |
unchanged | hooks |
|
tests/test_id_normalization_contract.py |
unchanged | build, extract, ids, mcp_ingest, symbol_resolution |
Skips as a whole: it needs hypothesis, which the pinned reference environment does not install |
tests/test_install_references.py |
changed | command line | The one test check_install selects from it always skips (every host bundle has shipped) |
tests/test_install_strings.py |
unchanged | command line | |
tests/test_install_upgrade.py |
unchanged | command line | |
tests/test_label_retry.py |
unchanged | llm |
|
tests/test_labeling.py |
unchanged | command line, llm, watch |
|
tests/test_llm_null_token_counts.py |
new | llm |
All 4 tests drive mocked Python provider SDKs, so check_llm deselects them; check_llm_execution compares the same contract on the engine's own transports |
tests/test_merge_chunks_validation.py |
changed | command line, semantic_cleanup |
|
tests/test_ollama.py |
unchanged | llm |
All its tests drive mocked Python provider SDKs, so check_llm deselects them |
tests/test_ollama_retry_cap.py |
changed | llm |
|
tests/test_openai_custom_endpoint.py |
unchanged | llm |
|
tests/test_oversized_document_slicing.py |
unchanged | detect, file_slice, llm |
|
tests/test_pdf_token_estimate.py |
unchanged | llm |
|
tests/test_pr10_watch_overrides.py |
new | watch (through tests/test_language_override_lifecycle.py) |
Skips as a whole: it needs watchdog, which the pinned reference environment does not install |
tests/test_provider_registry.py |
unchanged | llm |
|
tests/test_query_cli.py |
unchanged | command line | |
tests/test_settings_merge.py |
unchanged | install |
|
tests/test_skill_version_warning.py |
unchanged | command line | |
tests/test_uninstall_scope.py |
changed | install |