Skip to content

fix(gateway): reasoning display is silently lost on every streamed turn #57693

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
CocaKova wants to merge 1 commit into
base: main
Choose a base branch
from
+370 −51
Open

fix(gateway): reasoning display is silently lost on every streamed turn #57693

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
181 changes: 130 additions & 51 deletions gateway/run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10892,6 +10892,116 @@ def _get_cached_session_source(self, session_key: str):
pass
return source

def _format_reasoning_block(self, source, last_reasoning) -> str:
"""Format the model's reasoning as the displayable 💭 block, or "".

Shared by the normal send path and the streamed-turn reasoning fold:
resolves the per-platform ``show_reasoning`` switch (Mattermost
requires an explicit platform override because this is scratch text,
not ordinary final-answer content), collapses long reasoning to 15
lines, and renders the per-platform ``reasoning_style`` (code /
subtext / blockquote). Returns "" when display is off or there is no
reasoning to show.
"""
try:
_show_reasoning_effective = _resolve_gateway_display_bool(
_load_gateway_config(),
_platform_config_key(source.platform),
"show_reasoning",
default=bool(getattr(self, "_show_reasoning", False)),
platform=source.platform,
require_platform_override_for={Platform.MATTERMOST},
)
except Exception:
_show_reasoning_effective = (
False
if source.platform == Platform.MATTERMOST
else getattr(self, "_show_reasoning", False)
)
if not _show_reasoning_effective or not last_reasoning:
return ""
# Collapse long reasoning to keep messages readable
lines = last_reasoning.strip().splitlines()
if len(lines) > 15:
display_reasoning = "\n".join(lines[:15])
display_reasoning += f"\n_... ({len(lines) - 15} more lines)_"
else:
display_reasoning = last_reasoning.strip()
# Render style is per-platform: Discord defaults to "-# " subtext
# (native small grey metadata text); other platforms keep the fenced
# code block.
try:
from gateway.display_config import resolve_display_setting
_reasoning_style = resolve_display_setting(
_load_gateway_config(),
_platform_config_key(source.platform),
"reasoning_style",
"code",
)
except Exception:
_reasoning_style = "code"
if _reasoning_style == "subtext":
_quoted = "\n".join(
f"-# {ln}" if ln else "-#" for ln in display_reasoning.splitlines()
)
return f"-# 💭 Reasoning\n{_quoted}"
if _reasoning_style == "blockquote":
_quoted = "\n".join(
f"> {ln}" if ln else ">" for ln in display_reasoning.splitlines()
)
return f"> 💭 **Reasoning:**\n{_quoted}"
return f"💭 **Reasoning:**\n```\n{display_reasoning}\n```"

async def _fold_reasoning_into_streamed_message(
self,
*,
source,
stream_consumer,
final_text,
last_reasoning,
session_key=None,
) -> bool:
"""Fold the 💭 reasoning block into an already-streamed final message.

The streamed commit bypasses the normal send path — the only place the
reasoning block is prepended — so turning streaming on silently
disabled reasoning display for every model and platform. Re-attach it
with one final edit routed through the stream consumer's
metadata-aware edit path (``_edit_message``), so the edit still carries
the routing metadata Slack uses to pick the workspace client and
Telegram uses for topic/thread routing — a raw ``adapter.edit_message``
would drop it and a non-default Slack workspace would lose the edit.

Best-effort: a failed edit only loses the reasoning display, never the
answer, and never un-suppresses the send. Returns True when an edit was
issued.
"""
if stream_consumer is None or not final_text:
return False
try:
message_id = stream_consumer.message_id
except Exception:
message_id = None
if not message_id:
return False
try:
reasoning_block = self._format_reasoning_block(source, last_reasoning)
if not reasoning_block:
return False
await stream_consumer._edit_message(
message_id=message_id,
content=f"{reasoning_block}\n\n{final_text}",
finalize=True,
)
return True
except Exception as _reasoning_edit_err:
logger.warning(
"Failed to fold reasoning into streamed message for session %s: %s",
session_key or "?",
_reasoning_edit_err,
)
return False

async def _handle_message_with_agent(self, event, source, _quick_key: str, run_generation: int):
"""Inner handler that runs under the _running_agents sentinel guard."""
_msg_start_time = time.time()
Expand Down Expand Up @@ -11899,58 +12009,14 @@ async def _handle_message_with_agent(self, event, source, _quick_key: str, run_g
)

# Prepend reasoning/thinking if display is enabled (per-platform).
# Mattermost requires explicit per-platform opt-in because this is
# scratch text, not ordinary final-answer content.
try:
_show_reasoning_effective = _resolve_gateway_display_bool(
_load_gateway_config(),
_platform_config_key(source.platform),
"show_reasoning",
default=bool(getattr(self, "_show_reasoning", False)),
platform=source.platform,
require_platform_override_for={Platform.MATTERMOST},
# Formatting/enablement lives in _format_reasoning_block so the
# streamed-turn fold below reuses the exact same rendering.
if response and not _intentional_silence:
_reasoning_block = self._format_reasoning_block(
source, agent_result.get("last_reasoning")
)
except Exception:
_show_reasoning_effective = (
False
if source.platform == Platform.MATTERMOST
else getattr(self, "_show_reasoning", False)
)
if _show_reasoning_effective and response and not _intentional_silence:
last_reasoning = agent_result.get("last_reasoning")
if last_reasoning:
# Collapse long reasoning to keep messages readable
lines = last_reasoning.strip().splitlines()
if len(lines) > 15:
display_reasoning = "\n".join(lines[:15])
display_reasoning += f"\n_... ({len(lines) - 15} more lines)_"
else:
display_reasoning = last_reasoning.strip()
# Render style is per-platform: Discord defaults to "-# "
# subtext (native small grey metadata text); other
# platforms keep the fenced code block.
try:
from gateway.display_config import resolve_display_setting
_reasoning_style = resolve_display_setting(
_load_gateway_config(),
_platform_config_key(source.platform),
"reasoning_style",
"code",
)
except Exception:
_reasoning_style = "code"
if _reasoning_style == "subtext":
_quoted = "\n".join(
f"-# {ln}" if ln else "-#" for ln in display_reasoning.splitlines()
)
response = f"-# 💭 Reasoning\n{_quoted}\n\n{response}"
elif _reasoning_style == "blockquote":
_quoted = "\n".join(
f"> {ln}" if ln else ">" for ln in display_reasoning.splitlines()
)
response = f"> 💭 **Reasoning:**\n{_quoted}\n\n{response}"
else:
response = f"💭 **Reasoning:**\n```\n{display_reasoning}\n```\n\n{response}"
if _reasoning_block:
response = f"{_reasoning_block}\n\n{response}"

# Runtime-metadata footer — only on the FINAL message of the turn.
# Off by default (display.runtime_footer.enabled=false). When
Expand Down Expand Up @@ -20337,6 +20403,19 @@ def _stream_confirmed_final_delivery(
previewed=_previewed,
)
if not _is_empty_sentinel and not _transformed and (_streamed or _content_delivered):
# The streamed commit bypasses the normal send path (the only
# place the 💭 reasoning block is prepended), so before
# suppressing, fold the block into the already-streamed message
# with one final metadata-aware edit. Best-effort — a failed
# fold only loses the reasoning display, never the answer, and
# never un-suppresses the send.
await self._fold_reasoning_into_streamed_message(
source=source,
stream_consumer=_sc,
final_text=_final,
last_reasoning=response.get("last_reasoning"),
session_key=session_key,
)
logger.info(
"Suppressing normal final send for session %s: final delivery already confirmed (streamed=%s previewed=%s content_delivered=%s).",
session_key or "?",
Expand Down
Loading
Loading