[bug] [opencode] bad headers from any provider could result in negative 'max-retry'

[kennyjinhiro]

Description

SessionRetry.delay accepts negative values for retry-after and retry-after-ms.

For example, retry-after-ms: -1 returns -1.

Eventually, negative value then sadly reaches retry scheduling, instead of falling back to normal exponential backoff.

This is because if (retry-after) and if (retry-after-ms), are both nullable conditions for 0 only. Not negatives.

Expected

I will force negative retry values to use normal backoff instead.

Plugins

none

OpenCode version

1.18.15

Steps to reproduce

1. Call SessionRetry.delay(1, apiError({ "retry-after-ms": "-1" })).
2. Observe result is -1.
3. Repeat with retry-after: "-1"; result is -1000.

Screenshot and/or share link

Operating System

macOS 26.2, arm64

Terminal

Ghostty

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• infinite retry loop on Azure 429 with Retry-After: 0 #37076: infinite retry loop when Retry-After: 0 — same SessionRetry.delay() function, zero-value header edge case
• Retry backoff is uncapped when a retryable error has response headers but no retry-after #33728: uncapped exponential backoff when response headers exist but contain no retry-after — detailed analysis of the same function's header-branch logic

Your report covers the negative-value edge case in the same code path. It may be worth coordinating a fix with those issues.

[kennyjinhiro]

Hi @jlongster , this is my first opened-issue at an open-source.
I have been using OpenCode, and I hope this is an okay-ish triage that we could do.

Thanks!

[kennyjinhiro]

To note: The reported case is specifically negative numeric headers. Further improvements can be made for zero-header or missing-header behavior.