Code graph and memory for your coding agents, in one Rust program. For Claude Code, Codex, Cursor and any MCP client.
Upstream tests with known differences
On 14 October 2026 the comparison checks of the graph engine were widened: they now run most of the test files of Graphify 0.9.88 against the engine, some of them for the first time. A few of those files have known differences: tests that the engine fails. Each one is recorded below with the reason, deselected by its exact name in the check that runs its file, and listed in that check's receipt. They are not fixed in 0.1.0.
The files the checks do not run are listed in upstream-tests-not-run.md, and every file's status is in upstream-tests-0.9.88.json.
Known differences
| Upstream test | Check | Why the engine fails it |
|---|---|---|
tests/test_cli_export.py::test_cluster_only_warns_when_labeling_flags_are_ignored |
check_cli |
cluster-only with saved labels and --backend exits 0 without the reference's warning on stderr that the labeling flags are ignored. |
tests/test_cli_help.py::test_help_lists_prs_provider_and_export_formats |
check_cli |
The engine's --help text is its own: it does not list the prs command, it names the export formats in one list instead of export <format> lines, and it lists the hook-guard and hook-check internals that the reference hides. |
tests/test_extract_code_only_cli.py::test_code_only_force_preserves_existing_semantic_layer |
check_cli |
extract --force --code-only keeps the existing semantic layer (the test's checks of the graph pass), but the engine does not print the reference's existing semantic layer preserved line. |
tests/test_extract_code_only_cli.py::test_code_only_succeeds_without_key |
check_cli |
extract --code-only with no API key builds the code graph, but the engine does not print the reference's --code-only: skipping line. |
tests/test_extract_code_only_cli.py::test_extract_names_skipped_sensitive_files |
check_cli |
The engine's extract does not print the reference's line that names the files it skipped as potentially sensitive. |
tests/test_extract_code_only_cli.py::test_extract_usage_advertises_code_only |
check_cli |
extract with no path prints the engine's error must specify a path to scan or a --postgres DSN instead of the reference's usage text, so --code-only is not mentioned. |
tests/test_typescript_enum_members.py::test_case_only_sibling_members_do_not_duplicate_an_edge |
check_extract |
Two TypeScript enum members whose names differ only in case (Value, value) share one node id. The reference keeps one case_of edge to it; the engine also emits a second one from the later member's line. |
Files first run on 14 October 2026
Against Graphify 0.9.77, most of these files are unchanged; the table marks the new and changed ones. "Left out" names the tests of a file that its check does not run, with the reason; the known differences above are left out too.
| Test file | Since 0.9.77 | Check | Left out |
|---|---|---|---|
tests/test_builtin_global_type_refs.py |
unchanged | check_extract |
|
tests/test_cargo_missing_manifest.py |
unchanged | check_cli |
|
tests/test_case_sensitive_resolution.py |
unchanged | check_extract |
|
tests/test_cli_export.py |
changed | check_cli |
|
tests/test_cli_help.py |
unchanged | check_cli |
|
tests/test_community_labels_skill.py |
unchanged | check_export |
test_core_fragments_step5_reexport_with_curated_labels: reads Graphify's own skill fragments; the engine installs its own skill files; test_skill_step5_reexports_graph_json_with_curated_labels: reads Graphify's own skill text; the engine installs its own skill files |
tests/test_corrupt_graph_json.py |
unchanged | check_build |
|
tests/test_cpp_method_declarations.py |
unchanged | check_extract |
|
tests/test_cross_language_call_resolution.py |
unchanged | check_extract |
|
tests/test_csharp_call_site_generic_args.py |
unchanged | check_extract |
|
tests/test_csharp_field_generic_args.py |
unchanged | check_extract |
|
tests/test_csharp_generic_callsites.py |
unchanged | check_extract |
|
tests/test_csharp_member_nodes.py |
unchanged | check_extract |
|
tests/test_csharp_tuple_type_refs.py |
unchanged | check_extract |
|
tests/test_definition_file_portability.py |
unchanged | check_extract |
|
tests/test_duplicate_annotation_edges.py |
unchanged | check_extract |
|
tests/test_elixir_import_resolution.py |
unchanged | check_extract |
|
tests/test_elixir_unqualified_call_scope.py |
unchanged | check_extract |
|
tests/test_evidence_binding.py |
unchanged | check_llm |
12 tests, all but test_label_identifiers_helper: they call none of the helpers that check_llm compares with the engine, and it runs only the tests that call them |
tests/test_export_idempotent_writes.py |
unchanged | check_export |
test_to_wiki_rerun_does_not_rewrite_unchanged_articles, test_to_wiki_rerun_sweeps_orphaned_relabelled_article: check_wiki compares the engine's pages with the reference's but leaves the reference's files in place, so a rerun check of file times would test the reference; test_write_text_atomic_if_changed_skips_identical: checks the reference's write_text_atomic_if_changed helper, which no check routes to the engine |
tests/test_extract_cache_location.py |
unchanged | check_extract |
|
tests/test_extract_code_only_cli.py |
unchanged | check_cli |
|
tests/test_file_node_id_spec.py |
unchanged | check_extract |
|
tests/test_gemini_hook.py |
unchanged | check_cli |
|
tests/test_import_self_loops.py |
unchanged | check_extract |
|
tests/test_incremental.py |
unchanged | check_extract |
|
tests/test_ingest_url_type.py |
unchanged | check_ingest |
|
tests/test_inherited_field_receivers.py |
unchanged | check_extract |
|
tests/test_injection_sentinel_coverage.py |
unchanged | check_llm |
|
tests/test_issue_3405_python_resolution.py |
unchanged | check_extract |
|
tests/test_java_enum_members.py |
unchanged | check_extract |
|
tests/test_js_callback_calls.py |
unchanged | check_extract |
|
tests/test_js_destructured_export.py |
unchanged | check_extract |
|
tests/test_js_exported_fn_call_resolution.py |
unchanged | check_extract |
|
tests/test_js_module_level_anonymous_calls.py |
unchanged | check_extract |
|
tests/test_js_this_member_expressions.py |
unchanged | check_extract |
|
tests/test_jsx_component_usage_calls.py |
unchanged | check_extract |
|
tests/test_kotlin_grammar.py |
unchanged | check_extract |
|
tests/test_kotlin_object_literal.py |
unchanged | check_extract |
|
tests/test_long_path_hashing.py |
unchanged | check_detect |
test_os_path_adds_prefix_on_win32: it simulates Windows by patching Python's sys.platform; the engine uses the platform it runs on |
tests/test_lua_import.py |
unchanged | check_extract |
|
tests/test_lua_method_calls.py |
unchanged | check_extract |
|
tests/test_merge_graphs_cli.py |
unchanged | check_cli |
|
tests/test_no_dedup_flag.py |
unchanged | check_build |
|
tests/test_node_id_canonical.py |
unchanged | check_extract |
|
tests/test_non_regular_files.py |
changed | check_detect |
|
tests/test_non_string_node_ids.py |
unchanged | check_build |
|
tests/test_objc_field_table_remap.py |
unchanged | check_extract |
|
tests/test_obsidian_dangling_member.py |
unchanged | check_export |
|
tests/test_obsidian_filename_cap.py |
unchanged | check_export |
|
tests/test_out_dir_evidence.py |
unchanged | check_detect |
|
tests/test_owned_type_sources.py |
unchanged | check_extract |
|
tests/test_package_json_subpath_imports.py |
unchanged | check_extract |
|
tests/test_phantom_cross_package_call.py |
unchanged | check_extract |
|
tests/test_read_hook.py |
unchanged | check_cli |
|
tests/test_ruby_bare_self_send_calls.py |
unchanged | check_extract |
|
tests/test_rust_macro_rules.py |
unchanged | check_rust |
|
tests/test_scala_abstract_methods.py |
unchanged | check_extract |
|
tests/test_semantic_id_remap_root.py |
unchanged | check_build |
|
tests/test_skill_part_a_explicit_root.py |
new | check_extract |
test_part_a_passes_an_explicit_root, test_split_hosts_are_discovered: reads the skill files that Graphify's skill generator (tools/skillgen) renders; the engine ships its own skill files |
tests/test_skill_persisted_scan_options.py |
new | check_detect |
|
tests/test_sourceless_call_targets.py |
unchanged | check_extract |
|
tests/test_stat_index_multi_root.py |
unchanged | check_cache |
test_each_root_gets_its_own_stat_index_file_not_the_first_roots: it changes the output folder between two calls; check_cache rebuilds the engine's cache state for each call from the reference's globals, which do not hold the stat index's file path |
tests/test_swift_cross_file_calls.py |
unchanged | check_extract |
|
tests/test_swift_import_resolution.py |
unchanged | check_extract |
|
tests/test_ts_abstract_methods.py |
unchanged | check_extract |
|
tests/test_ts_decorators.py |
unchanged | check_extract |
|
tests/test_ts_generators.py |
unchanged | check_extract |
|
tests/test_ts_import_require.py |
unchanged | check_extract |
|
tests/test_ts_import_type_arguments.py |
unchanged | check_extract |
|
tests/test_ts_inheritance.py |
unchanged | check_extract |
|
tests/test_ts_namespace.py |
unchanged | check_extract |
|
tests/test_ts_parse_warning.py |
unchanged | check_extract |
|
tests/test_typescript_enum_members.py |
unchanged | check_extract |
|
tests/test_zero_node_no_cache.py |
unchanged | check_extract |
test_zero_node_result_not_cached_then_self_heals: forces a zero-node result by patching the reference's private _safe_extract_with_xaml_root; the engine has no Python seam to patch, so its extraction is unchanged |