| name | integration-tests |
| description | Write and run integration tests against a GenLayer environment. |
| allowed-tools | ["Bash","Read","Write","Edit"] |
Integration Tests
Run contracts against a real GenLayer environment (GLSim, Studio, or testnet) with full consensus validation.
Running Tests
gltest tests/integration/ -v -s
gltest tests/integration/ -v -s --network localnet
gltest tests/integration/ -v -s --network studionet
gltest tests/integration/ -v -s --network testnet_bradbury
Always use -v -s for visible output during development.
Test Pattern
from gltest import get_contract_factory
from gltest.assertions import tx_execution_succeeded
def test_full_flow():
factory = get_contract_factory("MyContract")
contract = factory.deploy(args=[])
tx_receipt = contract.set_data(args=["hello"]).transact()
assert tx_execution_succeeded(tx_receipt)
result = contract.get_data(args=[contract.address]).call()
assert result == "hello"
ACCEPTED and FINALIZED are transaction lifecycle states, not proof that
contract execution succeeded. A transaction can be accepted and finalized with
an execution error, and failed execution applies no state changes. For deploy
transactions, failed execution means no contract is created.
Always assert tx_execution_succeeded(receipt) before reading state, checking
schema/code, or treating a missing contract as an infrastructure issue.
Key Differences from Direct Mode
| Direct Mode | Integration Tests |
|---|
| Speed | ~30ms | ~seconds to minutes |
| Server required | No | Yes (GLSim, Studio, or testnet) |
| Consensus | Leader only | Full leader + validators |
| Write methods | Return values directly | Return transaction receipts |
| Read methods | Return values directly | Use .call() |
| Mocking | mock_web() / mock_llm() | Real web/LLM calls |
Write vs Read Calls
Write methods (state-changing):
tx_receipt = contract.method_name(args=[arg1, arg2]).transact()
assert tx_execution_succeeded(tx_receipt)
Read methods (view-only):
result = contract.view_method(args=[arg1]).call()
Configuration (gltest.config.yaml)
contract_path: contracts/
networks:
localnet:
studionet:
testnet_bradbury:
accounts:
- "${ACCOUNT_PRIVATE_KEY_1}"
- "${ACCOUNT_PRIVATE_KEY_2}"
Test Markers
import pytest
@pytest.mark.slow
def test_expensive_operation():
"""Excluded by default. Run with: gltest -m slow"""
pass
Environments
- GLSim (
pip install genlayer-test[sim], glsim --port 4000 --validators 5) — lightweight, no Docker, ~1s startup. Runs Python natively, not in GenVM. Good for fast iteration.
- Studio local (
genlayer up) — full GenVM, real consensus, Docker required. Validates runtime compatibility.
- studio.genlayer.com (StudioNet) — hosted Studio, no setup, rate-limited (see Common Issues). Gasless: no tokens required. Accounts with 0 GEN balance can deploy and run tests normally.
- Testnet Bradbury — real network, requires funded accounts.
When to Use Integration Tests
- Validating consensus (leader + validators agree)
- Testing real web requests and LLM calls
- Pre-deployment smoke tests
- Verifying contract works in actual GenVM (not just Python runner)
Direct mode should cover most logic testing. Use integration tests for final validation before deploying.
Common Issues
"Transaction not found" errors
Clear cache: rm -rf .gltest_cache
Slow tests
Run single tests during development:
gltest tests/integration/test_file.py::test_specific -v -s
JSON serialization
When working with mock validators, convert to dicts:
transaction_context = {"validators": [v.to_dict() for v in mock_validators]}
Studio rate limits (HTTP 429 / -32429)
studio.genlayer.com enforces per-IP limits: 60 req/min, 1000 req/hr, 10000 req/day. Limits aren't permanent — once tripped, further requests are rejected until the current window resets (next minute / hour / day cycle). Throttle batch tests, run heavy suites against localnet (GLSim or local Studio), or pace .transact() calls.
-32028 is the related pending-queue cap — up to 32 in-flight txs per sender; a separate cap also applies per contract to prevent flooding the shared Studio. Wait for receipts before submitting the next batch instead of firing in parallel.