Fix "Failed to parse grammar" error for requests without grammar field

[Copilot]

OpenAI-compatible clients sending chat completion requests without a grammar field receive "Failed to parse grammar" errors, blocking integration with tools like opencode.

Root Cause

Two issues in the grammar parameter handling:

1. Missing grammar field defaults to server's global config instead of empty string
2. Empty grammar strings passed to common_sampler_init cause nullptr returns

Changes

server-task.cpp line 281: Use empty string default instead of inheriting from server config

// Before: inherits from server configuration
params.sampling.grammar = json_value(data, "grammar", defaults.sampling.grammar);

// After: defaults to empty string
params.sampling.grammar = json_value(data, "grammar", std::string());

server-context.cpp lines 1058-1065: Clear grammar-related fields when grammar is empty before sampler initialization

// Clear grammar triggers and lazy flag when grammar string is empty
if (task.params.sampling.grammar.empty()) {
    task.params.sampling.grammar_triggers.clear();
    task.params.sampling.grammar_lazy = false;
}

This ensures common_sampler_init doesn't attempt to parse empty grammar strings while preserving behavior for requests with explicit grammar constraints.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Failed to parse grammar - opencode</issue_title>
<issue_description>When I try to connect opencode with the docker model I only get this answer:
Failed to parse grammar

My open code configuration is:

"dmr": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Docker Model Runner",
      "options": {
        "baseURL": "http://localhost:12434/engines/v1"
      },
      "models": {
        "ai/qwen3-coder": {
          "name": "Qwen3 Coder"
        }
      }
    }

I use a macbook pro and have installed Docker model with version: Docker Model Runner version v1.0.2

I also tried it using my new gigabyte AI TOP ATOM.

"dmr": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Docker Model Runner",
      "options": {
        "baseURL": "http://localhost:12434/engines/v1"
      },
      "models": {
        "ai/qwen3-coder": {
          "name": "Qwen3 Coder"
        }
      }
    }

With Docker Model Runner version v1.0.6

I followed the instruction based on: www.docker.com/blog/new-nvidia-spark-docker-model-runner/

I hope someone can help me out

</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Failed to parse grammar - opencode docker/model-runner#534

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.