fix: flush block streaming on paragraph boundaries for chunkMode=newline (#7014)

* feat: Implement paragraph boundary flushing in block streaming

- Added `flushOnParagraph` option to `BlockReplyChunking` for immediate flushing on paragraph breaks.
- Updated `EmbeddedBlockChunker` to handle paragraph boundaries during chunking.
- Enhanced `createBlockReplyCoalescer` to support flushing on enqueue.
- Added tests to verify behavior of flushing with and without `flushOnEnqueue` set.
- Updated relevant types and interfaces to include `flushOnParagraph` and `flushOnEnqueue` options.

* fix: Improve streaming behavior and enhance block chunking logic

- Resolved issue with stuck typing indicator after streamed BlueBubbles replies.
- Refactored `EmbeddedBlockChunker` to streamline fence-split handling and ensure maxChars fallback for newline chunking.
- Added tests to validate new chunking behavior, including handling of paragraph breaks and fence scenarios.
- Updated changelog to reflect these changes.

* test: Add test for clamping long paragraphs in EmbeddedBlockChunker

- Introduced a new test case to verify that long paragraphs are correctly clamped to maxChars when flushOnParagraph is enabled.
- Updated logic in EmbeddedBlockChunker to handle cases where the next paragraph break exceeds maxChars, ensuring proper chunking behavior.

* refactor: streamline logging and improve error handling in message processing

- Removed verbose logging statements from the `processMessage` function to reduce clutter.
- Enhanced error handling by using `runtime.error` for typing restart failures.
- Updated the `applySystemPromptOverrideToSession` function to accept a string directly instead of a function, simplifying the prompt application process.
- Adjusted the `runEmbeddedAttempt` function to directly use the system prompt override without invoking it as a function.
This commit is contained in:
Tyler Yust
2026-02-02 01:22:41 -08:00
committed by GitHub
parent 85cd55e22b
commit 9ef24fd400
14 changed files with 377 additions and 73 deletions

View File

@@ -7,7 +7,7 @@ import {
INTERNAL_MESSAGE_CHANNEL,
listDeliverableMessageChannels,
} from "../../utils/message-channel.js";
import { resolveTextChunkLimit, type TextChunkProvider } from "../chunk.js";
import { resolveChunkMode, resolveTextChunkLimit, type TextChunkProvider } from "../chunk.js";
const DEFAULT_BLOCK_STREAM_MIN = 800;
const DEFAULT_BLOCK_STREAM_MAX = 1200;
@@ -54,6 +54,8 @@ export type BlockStreamingCoalescing = {
maxChars: number;
idleMs: number;
joiner: string;
/** When true, the coalescer flushes the buffer on each enqueue (paragraph-boundary flush). */
flushOnEnqueue?: boolean;
};
export function resolveBlockStreamingChunking(
@@ -64,22 +66,23 @@ export function resolveBlockStreamingChunking(
minChars: number;
maxChars: number;
breakPreference: "paragraph" | "newline" | "sentence";
flushOnParagraph?: boolean;
} {
const providerKey = normalizeChunkProvider(provider);
const providerConfigKey = providerKey;
const providerId = providerKey ? normalizeChannelId(providerKey) : null;
const providerChunkLimit = providerId
? getChannelDock(providerId)?.outbound?.textChunkLimit
: undefined;
const textLimit = resolveTextChunkLimit(cfg, providerKey, accountId, {
const textLimit = resolveTextChunkLimit(cfg, providerConfigKey, accountId, {
fallbackLimit: providerChunkLimit,
});
const chunkCfg = cfg?.agents?.defaults?.blockStreamingChunk;
// Note: chunkMode="newline" used to imply splitting on each newline, but outbound
// delivery now treats it as paragraph-aware chunking (only split on blank lines).
// Block streaming should follow the same rule, so we do NOT special-case newline
// mode here.
// (chunkMode no longer alters block streaming behavior)
// When chunkMode="newline", the outbound delivery splits on paragraph boundaries.
// The block chunker should flush eagerly on \n\n boundaries during streaming,
// regardless of minChars, so each paragraph is sent as its own message.
const chunkMode = resolveChunkMode(cfg, providerConfigKey, accountId);
const maxRequested = Math.max(1, Math.floor(chunkCfg?.maxChars ?? DEFAULT_BLOCK_STREAM_MAX));
const maxChars = Math.max(1, Math.min(maxRequested, textLimit));
@@ -90,7 +93,12 @@ export function resolveBlockStreamingChunking(
chunkCfg?.breakPreference === "newline" || chunkCfg?.breakPreference === "sentence"
? chunkCfg.breakPreference
: "paragraph";
return { minChars, maxChars, breakPreference };
return {
minChars,
maxChars,
breakPreference,
flushOnParagraph: chunkMode === "newline",
};
}
export function resolveBlockStreamingCoalescing(
@@ -102,17 +110,20 @@ export function resolveBlockStreamingCoalescing(
maxChars: number;
breakPreference: "paragraph" | "newline" | "sentence";
},
opts?: { chunkMode?: "length" | "newline" },
): BlockStreamingCoalescing | undefined {
const providerKey = normalizeChunkProvider(provider);
const providerConfigKey = providerKey;
// Note: chunkMode="newline" is paragraph-aware in outbound delivery (blank-line splits),
// so block streaming should not disable coalescing or flush per single newline.
// Resolve the outbound chunkMode so the coalescer can flush on paragraph boundaries
// when chunkMode="newline", matching the delivery-time splitting behavior.
const chunkMode = opts?.chunkMode ?? resolveChunkMode(cfg, providerConfigKey, accountId);
const providerId = providerKey ? normalizeChannelId(providerKey) : null;
const providerChunkLimit = providerId
? getChannelDock(providerId)?.outbound?.textChunkLimit
: undefined;
const textLimit = resolveTextChunkLimit(cfg, providerKey, accountId, {
const textLimit = resolveTextChunkLimit(cfg, providerConfigKey, accountId, {
fallbackLimit: providerChunkLimit,
});
const providerDefaults = providerId
@@ -149,5 +160,6 @@ export function resolveBlockStreamingCoalescing(
maxChars,
idleMs,
joiner,
flushOnEnqueue: chunkMode === "newline",
};
}