verifiers: fix tool call rendering from saved outputs

verifiers/utils/eval_display.py

@@ -6,12 +6,10 @@
 - TUI mode (screen=True): Alternate screen buffer with echo handling
 """
 
-import json
 import time
-from collections.abc import Mapping
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Any, Literal
+from typing import Literal
 
 from rich.columns import Columns
 from rich.console import Group
@@ -22,6 +20,7 @@
 
 from verifiers.types import EvalConfig, GenerateOutputs
 from verifiers.utils.display_utils import BaseDisplay, make_aligned_row
+from verifiers.utils.message_utils import format_messages
 
 
 @dataclass
@@ -59,54 +58,6 @@ def elapsed_time(self) -> float:
         return end - self.start_time
 
 
-def _format_messages(messages: Any) -> Text:
-    """Format messages for display (similar to print_prompt_completions_sample)."""
-
-    def _attr_or_key(obj: Any, key: str, default: Any = None) -> Any:
-        val = getattr(obj, key, None)
-        if val is not None:
-            return val
-        if isinstance(obj, Mapping):
-            return obj.get(key, default)
-        return default
-
-    def _normalize_tool_call(tc: Any) -> dict[str, str]:
-        src = _attr_or_key(tc, "function") or tc
-        name = _attr_or_key(src, "name", "") or ""
-        args = _attr_or_key(src, "arguments", {}) or {}
-        if not isinstance(args, str):
-            try:
-                args = json.dumps(args)
-            except Exception:
-                args = str(args)
-        return {"name": name, "args": args}
-
-    if isinstance(messages, str):
-        return Text(messages)
-
-    out = Text()
-    for idx, msg in enumerate(messages):
-        if idx:
-            out.append("\n\n")
-
-        assert isinstance(msg, dict)
-        role = msg.get("role", "")
-        content = msg.get("content", "")
-        style = "bright_cyan" if role == "assistant" else "bright_magenta"
-
-        out.append(f"{role}: ", style="bold")
-        out.append(str(content) if content else "", style=style)
-
-        for tc in msg.get("tool_calls") or []:
-            payload = _normalize_tool_call(tc)
-            out.append(
-                "\n\n[tool call]\n" + json.dumps(payload, indent=2, ensure_ascii=False),
-                style=style,
-            )
-
-    return out
-
-
 def _make_histogram(values: list[float], bins: int = 10, width: int = 20) -> Text:
     """Create a simple text histogram of values."""
     if not values:
@@ -577,14 +528,14 @@ def _make_env_detail(
             # Prompt panel
             items.append(
                 Panel(
-                    _format_messages(prompt),
+                    format_messages(prompt),
                     title="[dim]example 0 — prompt[/dim]",
                     border_style="dim",
                 )
             )
 
             # Completion panel (with error if any)
-            completion_text = _format_messages(completion)
+            completion_text = format_messages(completion)
             if error_0 is not None:
                 completion_text.append("\n\nerror: ", style="bold red")
                 completion_text.append(error_0, style="bold red")

verifiers/utils/logging_utils.py

@@ -1,7 +1,5 @@
-import json
 import logging
 import sys
-from collections.abc import Mapping
 from contextlib import contextmanager
 
 from rich.console import Console
@@ -12,6 +10,7 @@
 from verifiers.errors import Error
 from verifiers.types import Messages
 from verifiers.utils.error_utils import ErrorChain
+from verifiers.utils.message_utils import format_messages
 
 LOGGER_NAME = "verifiers"
 
@@ -81,57 +80,6 @@ def print_prompt_completions_sample(
     step: int,
     num_samples: int = 1,
 ) -> None:
-    def _attr_or_key(obj, key: str, default=None):
-        """Return obj.key if present, else obj[key] if Mapping, else default."""
-        val = getattr(obj, key, None)
-        if val is not None:
-            return val
-        if isinstance(obj, Mapping):
-            return obj.get(key, default)
-        return default
-
-    def _normalize_tool_call(tc):
-        """Return {"name": ..., "args": ...} from a dict or Pydantic-like object."""
-        src = (
-            _attr_or_key(tc, "function") or tc
-        )  # prefer nested function object if present
-        name = _attr_or_key(src, "name", "") or ""
-        args = _attr_or_key(src, "arguments", {}) or {}
-
-        if not isinstance(args, str):
-            try:
-                args = json.dumps(args)
-            except Exception:
-                args = str(args)
-        return {"name": name, "args": args}
-
-    def _format_messages(messages) -> Text:
-        if isinstance(messages, str):
-            return Text(messages)
-
-        out = Text()
-        for idx, msg in enumerate(messages):
-            if idx:
-                out.append("\n\n")
-
-            assert isinstance(msg, dict)
-            role = msg.get("role", "")
-            content = msg.get("content", "")
-            style = "bright_cyan" if role == "assistant" else "bright_magenta"
-
-            out.append(f"{role}: ", style="bold")
-            out.append(content, style=style)
-
-            for tc in msg.get("tool_calls") or []:  # treat None as empty list
-                payload = _normalize_tool_call(tc)
-                out.append(
-                    "\n\n[tool call]\n"
-                    + json.dumps(payload, indent=2, ensure_ascii=False),
-                    style=style,
-                )
-
-        return out
-
     def _format_error(error: str | BaseException) -> Text:
         out = Text()
         if isinstance(error, str):
@@ -158,8 +106,8 @@ def _format_error(error: str | BaseException) -> Text:
         error = errors[i]
         reward = reward_values[i]
 
-        formatted_prompt = _format_messages(prompt)
-        formatted_completion = _format_messages(completion)
+        formatted_prompt = format_messages(prompt)
+        formatted_completion = format_messages(completion)
         if error is not None:
             formatted_completion += Text("\n\n") + _format_error(error)
 

verifiers/utils/message_utils.py

@@ -1,9 +1,11 @@
 import json
-from typing import cast
+from collections.abc import Mapping
+from typing import Any, cast
 
 from openai.types.chat import (
     ChatCompletionAssistantMessageParam,
 )
+from rich.text import Text
 
 from verifiers.types import ChatMessage, Messages
 
@@ -85,6 +87,55 @@ def messages_to_printable(messages: Messages) -> Messages:
     return [message_to_printable(m) for m in messages or []]
 
 
+def format_messages(messages: Any) -> Text:
+    def _attr_or_key(obj: Any, key: str, default: Any = None) -> Any:
+        val = getattr(obj, key, None)
+        if val is not None:
+            return val
+        if isinstance(obj, Mapping):
+            return obj.get(key, default)
+        return default
+
+
+    def _normalize_tool_call(tc: Any) -> dict[str, str]:
+        if isinstance(tc, str):
+            tc = json.loads(tc)
+        src = _attr_or_key(tc, "function") or tc
+        name = _attr_or_key(src, "name", "") or ""
+        args = _attr_or_key(src, "arguments", {}) or {}
+        if not isinstance(args, str):
+            try:
+                args = json.dumps(args)
+            except Exception:
+                args = str(args)
+        return {"name": name, "args": args}
+
+    if isinstance(messages, str):
+        return Text(messages)
+
+    out = Text()
+    for idx, msg in enumerate(messages):
+        if idx:
+            out.append("\n\n")
+
+        assert isinstance(msg, dict)
+        role = msg.get("role", "")
+        content = msg.get("content", "")
+        style = "bright_cyan" if role == "assistant" else "bright_magenta"
+
+        out.append(f"{role}: ", style="bold")
+        out.append(str(content) if content else "", style=style)
+
+        for tc in msg.get("tool_calls") or []:
+            payload = _normalize_tool_call(tc)
+            out.append(
+                "\n\n[tool call]\n" + json.dumps(payload, indent=2, ensure_ascii=False),
+                style=style,
+            )
+
+    return out
+
+
 def sanitize_tool_calls(messages: Messages):
     """
     Sanitize tool calls from messages.