Performance notes

Injex is designed for small explicit service graphs. Injex adds cached dependency plans and a fast resolve path for the common case: singleton infrastructure plus transient application services.

Benchmark shape

The benchmark resolves this graph repeatedly:

• singleton Settings instance;
• singleton ApiClient(settings);
• transient UserRepository(client);
• transient EmailSender(client);
• transient AuditLog(settings);
• transient RegisterUser(repository, email_sender, audit_log).

This mirrors a common service-layer shape: app-wide configuration and clients, with per-operation use cases.

Local result

Environment used for the project benchmark:

• Python 3.13.5;
• macOS arm64;
• injex 1.5.0;
• wireup 2.11.3;
• dishka 1.10.1;
• dependency-injector 4.49.1;
• lagom 2.7.7;
• punq 0.7.0.

Library	Median resolve time
manual wiring	0.266 µs/op
Injex	0.333 µs/op
dishka	0.786 µs/op
Wireup, same scope	0.872 µs/op
Wireup, scope per operation	1.544 µs/op
dependency-injector	1.709 µs/op
lagom	9.487 µs/op
punq	56.982 µs/op

These numbers are not a universal ranking. They are a small synthetic benchmark for one graph shape. Different lifetimes, framework integrations, factories, async resources, and request context models can change results. dishka in particular is measured here on a synchronous graph; its async-resource and scope features are not exercised, so treat its number as "this graph," not "dishka in general."

Reproduce

Run from the repository root:

uv run --with punq --with lagom --with dependency-injector --with wireup --with dishka \
  python benchmarks/resolve_graph.py

The benchmark prints Python/package versions, the graph shape, median time, min and max samples, and relative overhead compared with manual wiring.

See also: benchmarks/README.md.

What this means

The result supports Injex's niche: explicit typed wiring can stay small while keeping hot-path resolve overhead low. Use the numbers as a sanity check, not as a substitute for measuring your own application graph.