Support streaming SDT segments

This commit is contained in:
Abe Jellinek 2026-05-06 13:16:25 -04:00
parent 73f6264e8f
commit ecb18449be
2 changed files with 146 additions and 29 deletions

View file

@ -86,6 +86,23 @@ class PDFWorker {
});
}
// Like _query, but onPartial fires for each isPartial message before the
// terminal response resolves the promise. Returns { promise, abort }.
_streamingQuery(action, data, transfer, onPartial) {
this._lastPromiseID++;
let id = this._lastPromiseID;
let promise = new Promise((resolve, reject) => {
this._waitingPromises[id] = { resolve, reject, onPartial };
this._worker.postMessage({ id, action, data }, transfer);
});
let abort = () => {
if (this._worker && this._waitingPromises[id]) {
this._worker.postMessage({ action: 'abort', id });
}
};
return { promise, abort };
}
_init() {
if (this._worker) return;
this._worker = new Worker(WORKER_URL);
@ -93,11 +110,22 @@ class PDFWorker {
let message = event.data;
if ('responseID' in message) {
let promise = this._waitingPromises[message.responseID];
delete this._waitingPromises[message.responseID];
if (!promise) {
Zotero.debug(`Received response from PDF worker for unknown request ${message.responseID}`);
return;
}
if (message.isPartial) {
if (promise.onPartial) {
try {
promise.onPartial(message.data);
}
catch (e) {
Zotero.logError(e);
}
}
return;
}
delete this._waitingPromises[message.responseID];
let { resolve, reject } = promise;
if ('error' in message) {
reject(new Error(JSON.stringify(message.error)));
@ -669,48 +697,103 @@ class PDFWorker {
*/
async getStructuredData(itemID, isPriority, password) {
return this._enqueue(async () => {
let attachment = await Zotero.Items.getAsync(itemID);
let prep = await this._prepareStructuredDataRequest(itemID);
if (!prep) return null;
let { attachment, contentType, buf } = prep;
Zotero.debug(`Getting structured document text from item ${attachment.libraryKey}`);
let t = new Date();
let contentType = attachment.attachmentContentType;
if (!(attachment.isPDFAttachment()
|| attachment.isEPUBAttachment()
|| attachment.isSnapshotAttachment())) {
throw new Error('Item must be a PDF, EPUB, or snapshot attachment');
}
let path = await attachment.getFilePathAsync();
if (!path) {
return null;
}
let buf = await IOUtils.read(path);
buf = new Uint8Array(buf).buffer;
try {
var result = await this._query('getStructuredDocumentText', {
buf, contentType, password
}, [buf]);
}
catch (e) {
let error = new Error(`Worker action 'getStructuredDocumentText' failed: ${JSON.stringify({ error: e.message })}`);
try {
error.name = JSON.parse(e.message).name;
}
catch (e) {
Zotero.logError(e);
}
Zotero.logError(error);
throw error;
throw this._wrapStructuredDataError(e);
}
Zotero.debug(`Extracted structured document text for item ${attachment.libraryKey} in ${new Date() - t} ms`);
return result;
}, isPriority);
}
async _prepareStructuredDataRequest(itemID) {
let attachment = await Zotero.Items.getAsync(itemID);
if (!(attachment.isPDFAttachment()
|| attachment.isEPUBAttachment()
|| attachment.isSnapshotAttachment())) {
throw new Error('Item must be a PDF, EPUB, or snapshot attachment');
}
let path = await attachment.getFilePathAsync();
if (!path) return null;
let buf = await IOUtils.read(path);
return {
attachment,
contentType: attachment.attachmentContentType,
buf: new Uint8Array(buf).buffer,
};
}
_wrapStructuredDataError(e) {
let error = new Error(`Worker action 'getStructuredDocumentText' failed: ${JSON.stringify({ error: e.message })}`);
try {
error.name = JSON.parse(e.message).name;
}
catch (parseErr) {
Zotero.logError(parseErr);
}
Zotero.logError(error);
return error;
}
// Streaming variant of getStructuredData. onChunk receives partial chunks
// ({ kind: 'partial', pages, content, pageIndexOffset, contentIndexOffset,
// pageIndexRange, totalPageCount }) followed by a final chunk
// ({ kind: 'final', structure }). Returns { promise, abort }.
getStructuredDataStream(itemID, onChunk, options = {}) {
let abortFn = null;
let aborted = false;
let { password, batchSize, isPriority } = options;
let promise = this._enqueue(async () => {
if (aborted) {
let e = new Error('Aborted');
e.name = 'AbortError';
throw e;
}
let prep = await this._prepareStructuredDataRequest(itemID);
if (!prep) return;
let { attachment, contentType, buf } = prep;
Zotero.debug(`Streaming structured document text from item ${attachment.libraryKey}`);
let t = new Date();
try {
let { promise: queryPromise, abort } = this._streamingQuery(
'getStructuredDocumentText',
{
buf,
contentType,
password,
streaming: true,
...(batchSize ? { batchSize } : {}),
},
[buf],
onChunk
);
abortFn = abort;
if (aborted) abort();
await queryPromise;
}
catch (e) {
throw this._wrapStructuredDataError(e);
}
Zotero.debug(`Streamed structured document text for item ${attachment.libraryKey} in ${new Date() - t} ms`);
}, isPriority);
return {
promise,
abort: () => {
aborted = true;
if (abortFn) abortFn();
},
};
}
/**
* Get data for recognizer-server
*

View file

@ -644,6 +644,40 @@ class ReaderInstance {
}
});
},
getSDTStream: (password, onChunk, onStart) => {
return new this._iframeWindow.Promise((resolve, reject) => {
let { promise, abort } = Zotero.PDFWorker.getStructuredDataStream(
this._item.id,
(chunk) => {
try {
onChunk(Cu.cloneInto(chunk, this._iframeWindow));
}
catch (e) {
Zotero.logError(e);
}
},
{
password,
isPriority: true,
}
);
if (onStart) {
try {
onStart(Cu.exportFunction(() => abort(), this._iframeWindow));
}
catch (e) {
Zotero.logError(e);
}
}
promise.then(
() => resolve(),
(e) => {
Zotero.logError(e);
reject(new this._iframeWindow.Error(e.message));
}
);
});
},
}, this._iframeWindow, { cloneFunctions: true }));
this._resolveInitPromise();