Python: extract_range(preserve_pairs=True) reorders messages around a call/result pair (fix open as PR #14165, unlinked)
メンテナーはふだん 2 日以内に返信
まだ誰も着手していません。
評価
- 難易度
- 2/5
- 見積もり時間
- 1〜3時間
- 初心者へのやさしさ
- 25/100
調査の方向性
Start in python/semantic_kernel/contents/history_reducer/chat_history_reducer_utils.py at extract_range() and run the provided reproduction with preserve_pairs=True. Confirm that an interleaved message remains between the call and result in the extracted history, then add or update regression coverage so the extracted order matches the original history.
索引モデルが issue の本文から書いたものです。
説明
Describe the bug
extract_range() in python/semantic_kernel/contents/history_reducer/chat_history_reducer_utils.py (used by ChatHistorySummarizationReducer with preserve_pairs=True) scrambles the chronological order of the extracted messages whenever a function-call/result pair has other messages interleaved between them.
Current main:
extracted: list[ChatMessageContent] = []
i = 0
while i < len(sliced):
idx = sliced[i]
msg = history[idx]
...
if preserve_pairs and idx in pair_map:
paired_idx = pair_map[idx]
if start <= paired_idx < end:
...
extracted.append(msg)
if paired_idx > idx:
extracted.append(history[paired_idx])
if paired_idx in sliced:
sliced.remove(paired_idx)
...
When the loop reaches the earlier half of a pair (the call), it immediately appends the later half (the result) right behind it in extracted, then removes the result's index from sliced so it isn't visited again at its natural position. That is what moves the result out of its original place whenever anything sits between the call and the result.
To Reproduce
Ran against the actual installed semantic-kernel package (1.44.1; chat_history_reducer_utils.py confirmed byte-identical to the current main file before running):
from semantic_kernel.contents.chat_message_content import ChatMessageContent
from semantic_kernel.contents.function_call_content import FunctionCallContent
from semantic_kernel.contents.function_result_content import FunctionResultContent
from semantic_kernel.contents.utils.author_role import AuthorRole
from semantic_kernel.contents.history_reducer.chat_history_reducer_utils import extract_range
msg0 = ChatMessageContent(role=AuthorRole.USER, content="What's the weather in Paris and Tokyo?")
msg1 = ChatMessageContent(role=AuthorRole.ASSISTANT, items=[FunctionCallContent(id="call_paris", name="get_weather")])
msg2 = ChatMessageContent(role=AuthorRole.USER, content="also check the forecast for tomorrow")
msg3 = ChatMessageContent(role=AuthorRole.TOOL, items=[FunctionResultContent(id="call_paris", name="get_weather", result="15C")])
msg4 = ChatMessageContent(role=AuthorRole.ASSISTANT, content="Paris is 15C.")
history = [msg0, msg1, msg2, msg3, msg4]
out = extract_range(history, start=0, end=5, preserve_pairs=True)
for m in out:
print(m.role, m.content or [getattr(it, "id", None) for it in m.items])
Output:
AuthorRole.USER What's the weather in Paris and Tokyo?
AuthorRole.ASSISTANT ['call_paris']
AuthorRole.TOOL ['call_paris']
AuthorRole.USER also check the forecast for tomorrow
AuthorRole.ASSISTANT Paris is 15C.
The tool result (originally index 3, after the interleaved user message at index 2) is pulled forward to sit right after the call, and the interleaved user message is pushed after it - the extracted order no longer matches history[0:5].
Expected behavior
extract_range(history, 0, 5, preserve_pairs=True) should return the messages in their original order (minus anything actually filtered out), i.e. history[0], history[1], history[2], history[3], history[4] unchanged, since nothing here should be filtered.
Platform
- Language: Python
- Source: pip package
semantic-kernel==1.44.1, confirmed unchanged on currentmain - File:
python/semantic_kernel/contents/history_reducer/chat_history_reducer_utils.py, functionextract_range()
Additional context
There is already an open, unmerged fix for exactly this for this: PR #14165 ("Python: Fix extract_range reordering messages when preserving function call/result pairs"), open since 2026-07-18, which rewrites the loop as a single forward pass so messages are never moved out of position. It has no linked issue, which may be why it has sat without a maintainer review beyond the automated Copilot pass. Filing this so the fix has something to attach to.
Note: while reviewing that PR's diff I found the rewritten version still shares the same pair_map[cidx] = ridx construction as current main, which has a separate, still-open dict-collision problem when a call message contains more than one function call (multiple results mapping to the same call index) - reported separately since it's a distinct bug from this ordering issue and reproduces on both the current code and PR #14165's rewrite.
- 主要言語
- C#
- スター
- 28.6k
- フォーク
- 4.8k
- 平均マージ
- 13時間 24分
- マージ済み PR(30日)
- 11
環境構築
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
microsoft/semantic-kernel のほかの issue
-
python triage
難易度 2/5 1〜3時間 初心者へのやさしさ 85/100
microsoft/semantic-kernel#14491 · コメント 1 件 ·
メンテナーはふだん 2 日以内に返信
-
python triage
難易度 2/5 1〜3時間 初心者へのやさしさ 82/100
microsoft/semantic-kernel#14490 · コメント 1 件 ·
メンテナーはふだん 2 日以内に返信
-
Python: [Python] structured_outputs_transform reuses ChatHistory across calls (prompt pollution)オープンpython triage
難易度 2/5 1〜3時間 初心者へのやさしさ 88/100
microsoft/semantic-kernel#14483 · コメント 2 件 ·
メンテナーはふだん 2 日以内に返信
-
.NET python triage
難易度 2/5 1〜3時間 初心者へのやさしさ 74/100
microsoft/semantic-kernel#14482 · コメント 3 件 ·
メンテナーはふだん 2 日以内に返信
-
python triage
難易度 2/5 1〜3時間 初心者へのやさしさ 85/100
microsoft/semantic-kernel#14481 ·
メンテナーはふだん 2 日以内に返信
microsoft/semantic-kernel の issue をすべて見る
似ている issue
-
WPF: each page's `Title` overwrites the window title, and returning to a page does not restore itオープン
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
メンテナーはふだん 1 日以内に返信
-
agentic-workflows area/Docs partner/agentic-workflows
難易度 1/5 1時間未満 初心者へのやさしさ 82/100
メンテナーはふだん 1 日以内に返信
-
bug
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
jamesmontemagno/tiny-clips#378 · コメント 2 件 ·
メンテナーはふだん 1 日以内に返信
-
go 🏃 testing 🧪
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
valkey-io/valkey-glide#7239 ·
メンテナーはふだん 2 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
SubtitleEdit/subtitleedit#15462 ·
メンテナーはふだん 1 日以内に返信