`retryCount: 0` increases user-facing errors

[freeatnet]

The use of retryCount: 0¹² in bundler and paymaster transports in @zerodev/sdk results in increased user-facing errors.

Steps to reproduce:

1. Prepare a dev machine with a network throttler (e.g., Network Conditioner.
2. Set up a sample ZeroDev project.
3. Open an interface where you can submit a user operation.
4. Temporarily interrupt the internet connection (e.g., in Network Conditioner, set "100% Loss" condition).
5. Attempt to submit the userop.
6. Immediately after, resume internet connectivity.

Expected: The transport should retry on network errors.
Observed: An error is thrown immediately.

Triage:

• The cause of the issue is the retryCount: 0 in transport initialization¹². The default is 3.
• retryCount > 0 applies only to transport errors. If the request completed with an RPC error, shouldRetry will return false and the request will not be retried.
• Further, viem will obey a numeric Retry-After header value (although the backend may need to add that header to Access-Control-Request-Headers for it to be readable by client).

Footnotes

1. https://github.com/zerodevapp/sdk/blob/74a06c58ae718ce341038382fa7dd989b71a6ad4/packages/core/clients/kernelAccountClient.ts#L89-L102 ↩ ↩²

2. https://github.com/zerodevapp/sdk/blob/74a06c58ae718ce341038382fa7dd989b71a6ad4/packages/core/clients/paymasterClient.ts#L64-L78 ↩ ↩²