Skip to content

Commit 7e4f726

Browse files
author
Bertan Ari
committed
fix(vscode-lm): restore em-dashes mangled during extraction
1 parent 88ad6d8 commit 7e4f726

1 file changed

Lines changed: 5 additions & 5 deletions

File tree

‎src/api/providers/vscode-lm.ts‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -72,8 +72,8 @@ function convertToVsCodeLmTools(tools: OpenAI.Chat.ChatCompletionTool[]): vscode
7272
* Copilot's backend enforces its own context window and, for third-party `sendRequest` callers,
7373
* trims an over-window request in a way that is NOT tool-pair-aware: it can drop the assistant
7474
* message holding a `tool_use` while keeping the matching `tool_result`, after which Anthropic
75-
* rejects the request with "unexpected tool_use_id". To keep trimming on OUR side — where
76-
* pairing is preserved — we shrink oversized `tool_result` payloads before sending. Only
75+
* rejects the request with "unexpected tool_use_id". To keep trimming on OUR side — where
76+
* pairing is preserved — we shrink oversized `tool_result` payloads before sending. Only
7777
* `tool_result` text is truncated (never `tool_use`, assistant text, summaries, or environment
7878
* details), and only when the request would otherwise exceed the budget.
7979
*/
@@ -85,7 +85,7 @@ function convertToVsCodeLmTools(tools: OpenAI.Chat.ChatCompletionTool[]): vscode
8585
* the tool schemas, image placeholders, or per-message framing the backend adds, so it cannot give
8686
* the true total for the request we are about to send. It is also an async, per-call RPC, and the
8787
* budget is needed for every message on every turn. We therefore keep a character estimate here
88-
* and stay deliberately conservative — 3 chars/token rather than the ~4 typical of English —
88+
* and stay deliberately conservative — 3 chars/token rather than the ~4 typical of English —
8989
* because the token-dense JSON, logs, and code that dominate oversized tool results tokenize to
9090
* fewer characters per token than prose. Under-counting biases toward trimming too early, which is
9191
* recoverable; over-counting sends an over-window request, which is not.
@@ -160,7 +160,7 @@ export function middleOutTruncate(text: string, maxChars: number): string {
160160
const headLength = Math.ceil(keep / 2)
161161
const tailLength = keep - headLength
162162
let head = text.slice(0, headLength)
163-
// Don't end the head on a lone high surrogate — its low half is in the removed middle, and a lone
163+
// Don't end the head on a lone high surrogate — its low half is in the removed middle, and a lone
164164
// surrogate cannot be encoded as UTF-8 (the backend 400s the whole request). Drop the split half.
165165
if (head.length > 0 && (head.charCodeAt(head.length - 1) & 0xfc00) === 0xd800) {
166166
head = head.slice(0, -1)
@@ -206,7 +206,7 @@ function estimateContentChars(content: Anthropic.Messages.MessageParam["content"
206206

207207
/**
208208
* Shrinks oversized `tool_result` payloads (largest first, middle-out) until the conversation fits
209-
* `budgetChars`. Mutates the tool_result blocks of the supplied messages in place — callers pass a
209+
* `budgetChars`. Mutates the tool_result blocks of the supplied messages in place — callers pass a
210210
* cloned array (see `createMessage`) so stored history is never mutated. A no-op when the
211211
* conversation already fits.
212212
*/

0 commit comments

Comments
 (0)