fix(provider): Z.AI GLM overflow error not classified as context overflow, causing chain compaction

[darkhipo]

Description

When Z.AI GLM models (glm-5-turbo, glm-5, glm-5.1) exceed their 200K context window, they return an error that is not recognized by opencode's OVERFLOW_PATTERNS in packages/opencode/src/provider/error.ts. This causes the error to be classified as a generic APICallError instead of ContextOverflowError, preventing auto-compaction and resulting in an unrecoverable chain compaction loop.

Observed Behavior

1. Session context fills past 200K threshold (common with many MCP tools)
2. opencode attempts compaction
3. Z.AI returns: "tokens in request more than max tokens allowed"
4. opencode classifies this as generic error (not overflow)
5. No compaction occurs → session retries → same error → infinite loop
6. Session becomes permanently unusable without DB intervention

Expected Behavior

The error message "tokens in request more than max tokens allowed" should be classified as context overflow, triggering auto-compaction.

Error Text

tokens in request more than max tokens allowed

Root Cause

OVERFLOW_PATTERNS in packages/opencode/src/provider/error.ts does not include a pattern matching Z.AI GLM's overflow error message.

Proposed Fix

Add the following pattern to OVERFLOW_PATTERNS:

/tokens in request more than max tokens allowed/i,

Related Issues

• fix(provider): classify "tokens in request more than max tokens allowed" as context overflow #27519 — same class of bug (OpenAI-compatible providers using this error phrasing)
• Session permanently stuck after auto-compaction: post-compaction auto-trigger fires tool_use without tool_result error and is non-recoverable #27594 — sessions permanently stuck after compaction (same downstream consequence)
• Auto-compaction triggers at ~200k tokens instead of model's actual context window (1M) #15871 — hardcoded context window ignoring actual model limits
• OAuth/Codex GPT-5.5 behaves like ~400k context while GPT-5.4 shows 1M #24561 — context limit mismatch between metadata and actual provider limits

Environment

• OpenCode version: 1.14.48
• Provider: zai-coding-plan (GLM-5-Turbo)
• The GLM-5 family (glm-5, glm-5-turbo, glm-5.1) all share 200K context documentation

[github-actions]

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

• fix(provider): classify "tokens in request more than max tokens allowed" as context overflow #27519: Proposes the exact same fix — adding /tokens in request more than max tokens allowed/i to OVERFLOW_PATTERNS — for the same error message from OpenAI-compatible providers. The root cause, error text, and proposed regex are identical; the only difference is that fix(provider): classify "tokens in request more than max tokens allowed" as context overflow #27519 is scoped more broadly while this issue focuses on Z.AI GLM specifically.
• GLM / Vertex overflow errors with INVALID_ARGUMENT are not classified as context overflow #23713: GLM / Vertex overflow errors with INVALID_ARGUMENT not classified as context overflow — same provider family, same class of bug.

Please check whether your issue is already covered by #27519 before proceeding. If the fix in #27519 also resolves the Z.AI GLM case, consider adding a comment there with the Z.AI-specific context rather than tracking it separately.