perf(ai): prepare assistant replay blocks in one pass (#141419)

Reuse the classified text, thinking, and tool-call collections and avoid joining text when replay needs content parts. Preserve Unicode sanitation, reasoning signatures, tool order, and the existing wire representation.
This commit is contained in:
Peter Steinberger 2026-09-07 11:39:36 -07:00 • committed by GitHub
parent 69faebcd4b
commit 8efdc03f8c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 71 additions and 54 deletions

View file

@ -80,7 +80,7 @@ describe("convertMessages assistant text replay", () => {
});
it.each([false, true])(
"preserves sanitized block positions with thinking-as-text %s",
"preserves interleaved text, thinking, and tool replay with thinking-as-text %s",
(requiresThinkingAsText) => {
const assistant: AssistantMessage = {
role: "assistant",
@ -88,20 +88,44 @@ describe("convertMessages assistant text replay", () => {
provider: model.provider,
model: model.id,
content: [
{ type: "thinking", thinking: "reason\ud800" },
{ type: "thinking", thinking: " \t", thinkingSignature: "reasoning_text" },
{
type: "thinking",
thinking: "reason\ud800",
thinkingSignature: "reasoning_content",
},
{ type: "text", text: " \t" },
{ type: "text", text: "first\ud800" },
{ type: "text", text: "\udc00" },
{ type: "thinking", thinking: "next😀" },
{
type: "toolCall",
id: "call_lookup",
name: "lookup",
arguments: { query: "cats" },
thoughtSignature: '{"type":"reasoning.encrypted","data":"synthetic"}',
},
{ type: "thinking", thinking: "next😀", thinkingSignature: "reasoning_text" },
{ type: "text", text: "last😀" },
],
usage: emptyUsage,
stopReason: "stop",
stopReason: "toolUse",
timestamp: 2,
};
const converted = convertMessages(
model,
{ messages: [assistant] },
{
messages: [
assistant,
{
role: "toolResult",
toolCallId: "call_lookup",
toolName: "lookup",
content: [{ type: "text", text: "found" }],
isError: false,
timestamp: 3,
},
],
},
{ ...resolveOpenAICompletionsCompat(model), requiresThinkingAsText },
);
@ -116,7 +140,17 @@ describe("convertMessages assistant text replay", () => {
{ type: "text", text: "last😀" },
]
: "first\n\nlast😀",
...(!requiresThinkingAsText && { reasoning_content: "reason\ud800\nnext😀" }),
tool_calls: [
{
id: "call_lookup",
type: "function",
function: { name: "lookup", arguments: '{"query":"cats"}' },
},
],
reasoning_details: [{ type: "reasoning.encrypted", data: "synthetic" }],
},
{ role: "tool", content: "found", tool_call_id: "call_lookup" },
]);
},
);

View file

@ -15,7 +15,7 @@ import {
isImageWithMediaPayload,
} from "./providers/tool-result-text.js";
import type { ResolvedOpenAICompletionsCompat } from "./transports/openai-completions-compat.js";
import type { Context, Model, TextContent, ThinkingContent, ToolCall } from "./types.js";
import type { Context, Model, ThinkingContent, ToolCall } from "./types.js";
import { sanitizeSurrogates } from "./utils/sanitize-unicode.js";
import { stripSystemPromptCacheBoundary } from "./utils/system-prompt-cache-boundary.js";
@ -25,18 +25,6 @@ type ChatCompletionContentPartVideo = {
video_url: { url: string };
};
function isTextContentBlock(block: { type: string }): block is TextContent {
return block.type === "text";
}
function isThinkingContentBlock(block: { type: string }): block is ThinkingContent {
return block.type === "thinking";
}
function isToolCallBlock(block: { type: string }): block is ToolCall {
return block.type === "toolCall";
}
function sanitizeToolResultText(text: string, fallback: string): string {
const sanitized = sanitizeSurrogates(text);
return sanitized.trim().length > 0 ? sanitized : fallback;
@ -153,35 +141,36 @@ export function convertMessages(
content: compat.requiresAssistantAfterToolResult ? "" : null,
};
const assistantTexts = msg.content
.filter(isTextContentBlock)
.filter((block) => block.text.trim().length > 0)
.map((block) => sanitizeSurrogates(block.text));
// Separate content blocks are distinct utterances, so replay them the way
// the string-content flattener does rather than running them together.
const assistantText = assistantTexts.join("\n");
const nonEmptyThinkingBlocks = msg.content
.filter(isThinkingContentBlock)
.filter((block) => block.thinking.trim().length > 0);
if (nonEmptyThinkingBlocks.length > 0) {
if (compat.requiresThinkingAsText) {
const thinkingText = nonEmptyThinkingBlocks
.map((block) => sanitizeSurrogates(block.thinking))
.join("\n\n");
assistantMsg.content = [
{ type: "text", text: thinkingText },
...assistantTexts.map((text): ChatCompletionContentPartText => ({
type: "text",
text,
})),
];
} else {
// String content is the interoperable Chat Completions replay shape;
// content-part arrays make some compatible servers mirror JSON.
if (assistantText.length > 0) {
assistantMsg.content = assistantText;
}
const assistantTexts: string[] = [];
const nonEmptyThinkingBlocks: ThinkingContent[] = [];
const toolCalls: ToolCall[] = [];
msg.content.forEach((block) => {
if (block.type === "text" && block.text.trim().length > 0) {
assistantTexts.push(sanitizeSurrogates(block.text));
} else if (block.type === "thinking" && block.thinking.trim().length > 0) {
nonEmptyThinkingBlocks.push(block);
} else if (block.type === "toolCall") {
toolCalls.push(block);
}
});
if (nonEmptyThinkingBlocks.length > 0 && compat.requiresThinkingAsText) {
const thinkingText = nonEmptyThinkingBlocks
.map((block) => sanitizeSurrogates(block.thinking))
.join("\n\n");
assistantMsg.content = [
{ type: "text", text: thinkingText },
...assistantTexts.map((text): ChatCompletionContentPartText => ({
type: "text",
text,
})),
];
} else {
// Keep separate utterances apart in the interoperable string replay shape.
const assistantText = assistantTexts.join("\n");
if (assistantText.length > 0) {
assistantMsg.content = assistantText;
}
if (nonEmptyThinkingBlocks.length > 0) {
let signature = nonEmptyThinkingBlocks.at(0)?.thinkingSignature;
if (model.provider === "opencode-go" && signature === "reasoning") {
signature = "reasoning_content";
@ -191,11 +180,8 @@ export function convertMessages(
nonEmptyThinkingBlocks.map((block) => block.thinking).join("\n");
}
}
} else if (assistantText.length > 0) {
assistantMsg.content = assistantText;
}
const toolCalls = msg.content.filter(isToolCallBlock);
if (toolCalls.length > 0) {
assistantMsg.tool_calls = toolCalls.map((toolCall) => ({
id: toolCall.id,
@ -231,10 +217,7 @@ export function convertMessages(
(assistantMsg as { reasoning_content?: string }).reasoning_content = "";
}
const content = assistantMsg.content;
const hasContent =
content !== null &&
content !== undefined &&
(typeof content === "string" ? content.length > 0 : content.length > 0);
const hasContent = content !== null && content !== undefined && content.length > 0;
if (!hasContent && !assistantMsg.tool_calls) {
continue;
}