ref(ai): Revert binary blob truncation

Author: alexander-alderman-webb

sentry_sdk/_types.py

@@ -12,7 +12,6 @@
 
 
 SENSITIVE_DATA_SUBSTITUTE = "[Filtered]"
-BLOB_DATA_SUBSTITUTE = "[Blob substitute]"
 
 
 class AnnotatedValue:

sentry_sdk/ai/utils.py

@@ -3,8 +3,6 @@
 from copy import deepcopy
 from typing import TYPE_CHECKING
 
-from sentry_sdk._types import BLOB_DATA_SUBSTITUTE
-from sentry_sdk.ai.consts import DATA_URL_BASE64_REGEX
 
 if TYPE_CHECKING:
     from typing import Any, Callable, Dict, List, Optional, Tuple
@@ -198,104 +196,6 @@ def _find_truncation_index(messages: "List[Dict[str, Any]]", max_bytes: int) ->
     return 0
 
 
-def _is_image_type_with_blob_content(item: "Dict[str, Any]") -> bool:
-    """
-    Some content blocks contain an image_url property with base64 content as its value.
-    This is used to identify those while not leading to unnecessary copying of data when the image URL does not contain base64 content.
-    """
-    if item.get("type") != "image_url":
-        return False
-
-    image_url = item.get("image_url", {}).get("url", "")
-    data_url_match = DATA_URL_BASE64_REGEX.match(image_url)
-
-    return bool(data_url_match)
-
-
-def redact_blob_message_parts(
-    messages: "List[Dict[str, Any]]",
-) -> "List[Dict[str, Any]]":
-    """
-    Redact blob message parts from the messages by replacing blob content with "[Filtered]".
-
-    This function creates a deep copy of messages that contain blob content to avoid
-    mutating the original message dictionaries. Messages without blob content are
-    returned as-is to minimize copying overhead.
-
-    e.g:
-    {
-        "role": "user",
-        "content": [
-            {
-                "text": "How many ponies do you see in the image?",
-                "type": "text"
-            },
-            {
-                "type": "blob",
-                "modality": "image",
-                "mime_type": "image/jpeg",
-                "content": "data:image/jpeg;base64,..."
-            }
-        ]
-    }
-    becomes:
-    {
-        "role": "user",
-        "content": [
-            {
-                "text": "How many ponies do you see in the image?",
-                "type": "text"
-            },
-            {
-                "type": "blob",
-                "modality": "image",
-                "mime_type": "image/jpeg",
-                "content": "[Filtered]"
-            }
-        ]
-    }
-    """
-
-    # First pass: check if any message contains blob content
-    has_blobs = False
-    for message in messages:
-        if not isinstance(message, dict):
-            continue
-        content = message.get("content")
-        if isinstance(content, list):
-            for item in content:
-                if isinstance(item, dict) and (
-                    item.get("type") == "blob" or _is_image_type_with_blob_content(item)
-                ):
-                    has_blobs = True
-                    break
-        if has_blobs:
-            break
-
-    # If no blobs found, return original messages to avoid unnecessary copying
-    if not has_blobs:
-        return messages
-
-    # Deep copy messages to avoid mutating the original
-    messages_copy = deepcopy(messages)
-
-    # Second pass: redact blob content in the copy
-    for message in messages_copy:
-        if not isinstance(message, dict):
-            continue
-
-        content = message.get("content")
-        if isinstance(content, list):
-            for item in content:
-                if isinstance(item, dict):
-                    if item.get("type") == "blob":
-                        item["content"] = BLOB_DATA_SUBSTITUTE
-                    elif _is_image_type_with_blob_content(item):
-                        item["image_url"]["url"] = BLOB_DATA_SUBSTITUTE
-
-    return messages_copy
-
-
 def truncate_messages_by_size(
     messages: "List[Dict[str, Any]]",
     max_bytes: int = MAX_GEN_AI_MESSAGE_BYTES,
@@ -341,8 +241,6 @@ def truncate_and_annotate_messages(
     if not messages:
         return None
 
-    messages = redact_blob_message_parts(messages)
-
     truncated_message = _truncate_single_message_content_if_present(
         deepcopy(messages[-1]), max_chars=max_single_message_chars
     )
@@ -361,8 +259,6 @@ def truncate_and_annotate_embedding_inputs(
     if not messages:
         return None
 
-    messages = redact_blob_message_parts(messages)
-
     truncated_messages, removed_count = truncate_messages_by_size(messages, max_bytes)
     if removed_count > 0:
         scope._gen_ai_original_message_count[span.span_id] = len(messages)