Merge pull request #13977 from Snuffleupagus/enqueueChunk-batch
[api-minor] Reduce `postMessage` overhead, in `PartialEvaluator.getTextContent`, by sending text chunks in batches (issue 13962)
This commit is contained in:
commit
e97f01b17c
@ -107,6 +107,17 @@ const PatternType = {
|
||||
SHADING: 2,
|
||||
};
|
||||
|
||||
// Optionally avoid sending individual, or very few, text chunks to reduce
|
||||
// `postMessage` overhead with ReadableStream (see issue 13962).
|
||||
//
|
||||
// PLEASE NOTE: This value should *not* be too large (it's used as a lower limit
|
||||
// in `enqueueChunk`), since that would cause streaming of textContent to become
|
||||
// essentially useless in practice by sending all (or most) chunks at once.
|
||||
// Also, a too large value would (indirectly) affect the main-thread `textLayer`
|
||||
// building negatively by forcing all textContent to be handled at once, which
|
||||
// could easily end up hurting *overall* performance (e.g. rendering as well).
|
||||
const TEXT_CHUNK_BATCH_SIZE = 10;
|
||||
|
||||
const deferred = Promise.resolve();
|
||||
|
||||
// Convert PDF blend mode names to HTML5 blend mode names.
|
||||
@ -2575,8 +2586,6 @@ class PartialEvaluator {
|
||||
if (textContentItem.initialized) {
|
||||
textContentItem.hasEOL = true;
|
||||
flushTextContentItem();
|
||||
} else if (textContent.items.length > 0) {
|
||||
textContent.items[textContent.items.length - 1].hasEOL = true;
|
||||
} else {
|
||||
textContent.items.push({
|
||||
str: "",
|
||||
@ -2658,20 +2667,24 @@ class PartialEvaluator {
|
||||
textContentItem.str.length = 0;
|
||||
}
|
||||
|
||||
function enqueueChunk() {
|
||||
function enqueueChunk(batch = false) {
|
||||
const length = textContent.items.length;
|
||||
if (length > 0) {
|
||||
sink.enqueue(textContent, length);
|
||||
textContent.items = [];
|
||||
textContent.styles = Object.create(null);
|
||||
if (length === 0) {
|
||||
return;
|
||||
}
|
||||
if (batch && length < TEXT_CHUNK_BATCH_SIZE) {
|
||||
return;
|
||||
}
|
||||
sink.enqueue(textContent, length);
|
||||
textContent.items = [];
|
||||
textContent.styles = Object.create(null);
|
||||
}
|
||||
|
||||
const timeSlotManager = new TimeSlotManager();
|
||||
|
||||
return new Promise(function promiseBody(resolve, reject) {
|
||||
const next = function (promise) {
|
||||
enqueueChunk();
|
||||
enqueueChunk(/* batch = */ true);
|
||||
Promise.all([promise, sink.ready]).then(function () {
|
||||
try {
|
||||
promiseBody(resolve, reject);
|
||||
|
@ -737,8 +737,6 @@ class WorkerMessageHandler {
|
||||
|
||||
handler.on("GetTextContent", function wphExtractText(data, sink) {
|
||||
const pageIndex = data.pageIndex;
|
||||
sink.onPull = function (desiredSize) {};
|
||||
sink.onCancel = function (reason) {};
|
||||
|
||||
pdfManager.getPage(pageIndex).then(function (page) {
|
||||
const task = new WorkerTask("GetTextContent: page " + pageIndex);
|
||||
|
@ -448,7 +448,7 @@ class MessageHandler {
|
||||
}
|
||||
// Reset desiredSize property of sink on every pull.
|
||||
this.streamSinks[streamId].desiredSize = data.desiredSize;
|
||||
const { onPull } = this.streamSinks[data.streamId];
|
||||
const { onPull } = this.streamSinks[streamId];
|
||||
new Promise(function (resolve) {
|
||||
resolve(onPull && onPull());
|
||||
}).then(
|
||||
@ -518,7 +518,7 @@ class MessageHandler {
|
||||
if (!this.streamSinks[streamId]) {
|
||||
break;
|
||||
}
|
||||
const { onCancel } = this.streamSinks[data.streamId];
|
||||
const { onCancel } = this.streamSinks[streamId];
|
||||
new Promise(function (resolve) {
|
||||
resolve(onCancel && onCancel(wrapReason(data.reason)));
|
||||
}).then(
|
||||
|
Loading…
x
Reference in New Issue
Block a user