Library API

Building a graph

from callix import PythonAdapter

adapter = PythonAdapter()
graph = adapter.analyze("path/to/project")

analyze() accepts an explicit file list, which skips discovery:

files = adapter.collect_files(root)
graph = adapter.analyze(root, files=[f for f in files if "tests" not in f.parts])

And a strict mode, which refuses to hand back a partial result:

graph = adapter.analyze(root, strict=True)   # AdapterError unless status is 'ok'

Adapter options

Every adapter takes the same four keyword-only arguments, so code that switches languages does not have to switch shape:

Adapter(resolve=False)                     # structure only
Adapter(resolver=my_resolver)              # replaces the native backend
Adapter(dep_parsers=[my_parser])           # replaces the manifest reader
Adapter(boundary_extractors=[my_finder])   # runs alongside the built-in ones

YamlAdapter takes resolve and boundary_extractors only: YAML declares no symbols, so there is nothing for a resolver to answer.

See Custom resolvers and parsers for the protocols and for what each adapter's native backend is.

Merging graphs

Analysing a polyglot repository means one graph per language, combined:

from callix import GoAdapter, PythonAdapter

graph = PythonAdapter().analyze(root)
graph.merge(GoAdapter().analyze(root), allow_shared=True)

allow_shared=True permits identical nodes to coincide — which is exactly what happens at a boundary, where two adapters deliberately produce the same BOUNDARY node. A collision between two different nodes with one ID stays an error even then.

Resolver statuses merge to the worst of the two, so a degraded half cannot hide behind a healthy one.

Serialization

data = graph.to_dict()          # JSON-compatible dict
text = graph.to_json(indent=2)

graph = Graph.from_dict(data)
graph = Graph.from_json(text)

Round-tripping is lossless, spans included. See Serialization for the schema.

Diffing

diff = old.diff(new)
diff.is_empty
diff.added_nodes, diff.removed_nodes, diff.changed_nodes
diff.added_relations, diff.removed_relations

Ordering inside each list is deterministic: nodes by ID, relations by their key. Edge metadata is part of that key, so two CALLS between the same pair from different call sites stay distinct, and a change to an edge's metadata alone still shows up.