@lynn123411/dsh-chat-translate 3.0.1 → 3.2.0

package/lib/types/server/pipeline/mask-tokens.d.ts

@@ -1,64 +0,0 @@
-/**
- * Mask placeholder tokens — single source of truth for the wire format, the
- * matcher and the leftover detector.
- *
- * Wire format: `⟦xkbdt3⟧` — U+27E6/U+27E7 (mathematical white square brackets)
- * around `<4-letter random id><index>`.
- *
- * Design constraints, each one a measured failure against real MT engines:
- *
- *  - No letters spelling a pronounceable word and no underscores: engines
- *    abbreviate tokens with such a shape back to their recognizable core, and
- *    bare `DSH` runs then show in the UI instead of the protected fragments.
- *  - Random letters keep two mask passes in the same session from colliding and
- *    make the token unambiguous in the translated text.
- *  - `matchMaskToken` accepts exactly what this module emits (plus the same
- *    token with a dropped closing bracket, an engine rewrite seen in practice).
- *    It deliberately does NOT accept anything else: a translation that damaged
- *    a token beyond recognition is discarded rather than repaired, because
- *    repairing it is what put mixed or duplicated text on screen.
- */
-export declare function randomTokenId(): string;
-export interface MaskTokenFormat {
-    /** Unique per `mask()` call; embedded in every token of that call. */
-    id: string;
-    /** Build the placeholder written into the text sent to the translator. */
-    token: (index: number) => string;
-}
-export declare function createMaskTokenFormat(): MaskTokenFormat;
-/**
- * Scanner for complete tokens of this format. A token whose closing bracket the
- * engine dropped is still resolved (see `matchMaskToken`), but it is not
- * recognized as a complete token here: the space `mask()` inserts sits outside
- * that token, and matching a truncated one would strip a space of the
- * translation's own.
- */
-export declare const MASK_TOKEN_PATTERN_SOURCE = "\u27E6([a-z]{4})(\\d+)\u27E7";
-/**
- * The opening bracket, the id and the index of a token, without its closing
- * bracket. Used to resolve a token the engine truncated.
- */
-export declare const MASK_TOKEN_PREFIX_PATTERN_SOURCE = "\u27E6\\s*([a-z]{4})(\\d+)";
-export interface MaskTokenMatch {
-    /** Index into the mask list of the `mask()` call that produced the token. */
-    index: number;
-    /** Length of the matched token in characters. */
-    length: number;
-    /** True when the closing bracket was missing and only the prefix matched. */
-    truncated: boolean;
-}
-/**
- * Match exactly one mask token of this format at `start` in `text`.
- *
- * The id must match the id of the masking pass that owns the token: every mask
- * pass carries its own random id, so a token carrying another id belongs to
- * another call and must not be resolved here.
- */
-export declare function matchMaskToken(text: string, start: number, id: string): MaskTokenMatch | null;
-/** Every token of this format in `text`, left to right. */
-export declare function findMaskTokens(text: string): Array<{
-    match: MaskTokenMatch;
-    raw: string;
-}>;
-/** True when `text` still carries a token of the current format. */
-export declare function hasMaskResidue(text: string): boolean;

package/lib/types/server/pipeline/masking.d.ts

@@ -1,45 +0,0 @@
-import { hasMaskResidue } from './mask-tokens.ts';
-export interface MaskResult {
-    maskedText: string;
-    /**
-     * Restore every protected fragment. Throws `MaskRestoreError` when the
-     * translated text did not carry the token sequence back intact; the caller
-     * must discard that translation instead of showing a repaired guess.
-     */
-    unmask: (translatedText: string) => string;
-}
-/** Which spaces `mask()` inserted directly before and after a token. */
-export interface InsertedSpacing {
-    leading: boolean;
-    trailing: boolean;
-}
-export declare class MaskRestoreError extends Error {
-    readonly expectedCount: number;
-    readonly foundCount: number;
-    constructor(message: string, expectedCount: number, foundCount: number);
-}
-export declare class ContentMaskingPipeline {
-    mask(text: string): MaskResult;
-}
-/**
- * Restore every token of `id` in `text` from `masks`.
- *
- * Each fragment of the pass is resolved exactly once, no matter how the engine
- * reordered the sentence: the index embedded in the token identifies the
- * fragment, and the random id identifies the masking pass that owns it. The
- * spaces recorded in `inserted` are removed together with their token, so the
- * spacing the translation produced around the fragment survives untouched.
- *
- * @throws MaskRestoreError when the text does not carry every token exactly
- *   once, or still shows a token of some other pass or a damaged one — the
- *   caller discards such a translation instead of rendering a repaired guess.
- */
-export declare function replaceMaskTokens(text: string, id: string, masks: string[], inserted?: InsertedSpacing[]): string;
-/**
- * True when a translated string still shows a mask token to the user. Callers
- * use it to discard a poisoned translation, to keep it out of the caches, and
- * to drop poisoned entries when a cache document is loaded (the file on disk
- * outlives the process and may carry anything).
- */
-export declare function isMaskLeak(translatedText: string): boolean;
-export { hasMaskResidue };