Stirling-PDF/shared-operations/workflow/traverseOperations.js
2023-11-10 21:08:07 +03:00

218 lines
8.1 KiB
JavaScript

import { organizeWaitOperations } from "./organizeWaitOperations.js";
import { extractPages } from "../functions/extractPages.js";
import { impose } from '../functions/impose.js';
import { mergePDFs } from '../functions/mergePDFs.js';
import { rotatePages } from '../functions/rotatePages.js';
import { scaleContent} from '../functions/scaleContent.js';
import { scalePage } from '../functions/scalePage.js';
import { splitPDF } from '../functions/splitPDF.js';
import { Metadata as dependantEditMetadata } from '../functions/editMetadata.js';
import { organizePages } from '../functions/organizePages.js';
import { removeBlankPages} from '../functions/removeBlankPages.js';
import { splitOn } from "../functions/splitOn.js";
/**
* @typedef PDF
* @property {string} originalFileName
* @property {string} fileName
* @property {Uint8Array} buffer
*/
/**
*
* @param {JSON} operations
* @param {PDF|PDF[]} input
* @returns {}
*/
export async function * traverseOperations(operations, input) {
const waitOperations = organizeWaitOperations(operations);
/** @type {PDF[]} */ let results = [];
yield* nextOperation(operations, input);
return results;
/**
*
* @param {JSON} operations
* @param {PDF|PDF[]} input
* @returns {undefined}
*/
async function * nextOperation(operations, input) {
if(Array.isArray(operations) && operations.length == 0) { // isEmpty
if(Array.isArray(input)) {
console.log("operation done: " + input[0].fileName + input.length > 1 ? "+" : "");
results = results.concat(input);
return;
}
else {
console.log("operation done: " + input.fileName);
results.push(input);
return;
}
}
for (let i = 0; i < operations.length; i++) {
yield* computeOperation(operations[i], structuredClone(input));
}
}
/**
*
* @param {JSON} operation
* @param {PDF|PDF[]} input
* @returns {undefined}
*/
async function * computeOperation(operation, input) {
yield "Starting: " + operation.type;
switch (operation.type) {
case "done": // Skip this, because it is a valid node.
break;
case "wait":
const waitOperation = waitOperations[operation.values.id];
if(Array.isArray(input)) {
waitOperation.input.concat(input); // TODO: May have unexpected concequences. Needs further testing!
}
else {
waitOperation.input.push(input);
}
waitOperation.waitCount--;
if(waitOperation.waitCount == 0) {
yield* nextOperation(waitOperation.doneOperation.operations, waitOperation.input);
}
break;
case "extract":
yield* nToN(input, operation, async (input) => {
input.fileName += "_extractedPages";
input.buffer = await extractPages(input.buffer, operation.values["pagesToExtractArray"]);
});
break;
case "impose":
yield* nToN(input, operation, async (input) => {
input.fileName += "_imposed";
input.buffer = await impose(input.buffer, operation.values["nup"], operation.values["format"]);
});
break;
case "merge":
yield* nToOne(input, operation, async (inputs) => {
return {
originalFileName: inputs.map(input => input.originalFileName).join("_and_"),
fileName: inputs.map(input => input.fileName).join("_and_") + "_merged",
buffer: await mergePDFs(inputs.map(input => input.buffer))
}
});
break;
case "rotate":
yield* nToN(input, operation, async (input) => {
input.fileName += "_turned";
input.buffer = await rotatePages(input.buffer, operation.values["rotation"]);
});
break;
case "split":
// TODO: A split might break the done condition, it may count multiple times. Needs further testing!
yield* oneToN(input, operation, async (input) => {
const splitResult = await splitPDF(input.buffer, operation.values["pagesToSplitAfterArray"]);
const splits = [];
for (let j = 0; j < splitResult.length; j++) {
splits.push({
originalFileName: input.originalFileName,
fileName: input.fileName + "_split" + j,
buffer: splitResult[j]
})
}
return splits;
});
break;
case "editMetadata":
yield* nToN(input, operation, async (input) => {
input.fileName += "_metadataEdited";
input.buffer = await editMetadata(input.buffer, operation.values["metadata"]);
});
break;
case "organizePages":
yield* nToN(input, operation, async (input) => {
input.fileName += "_pagesOrganized";
input.buffer = await organizePages(input.buffer, operation.values["operation"], operation.values["customOrderString"]);
});
break;
case "removeBlankPages":
yield* nToN(input, operation, async (input) => {
input.fileName += "_removedBlanks";
input.buffer = await removeBlankPages(input.buffer, operation.values["whiteThreashold"]);
});
break;
case "splitOn":
yield* oneToN(input, operation, async (input) => {
const splitResult = await splitOn(input.buffer, operation.values["type"], operation.values["whiteThreashold"]);
const splits = [];
for (let j = 0; j < splitResult.length; j++) {
splits.push({
originalFileName: input.originalFileName,
fileName: input.fileName + "_split" + j,
buffer: splitResult[j]
})
}
return splits;
});
break;
default:
throw new Error(`${operation.type} not implemented yet.`);
break;
}
}
/**
*
* @param {PDF|PDF[]} input
* @param {JSON} operation
* @returns {undefined}
*/
async function * nToOne(inputs, operation, callback) {
inputs = Array.from(inputs); // Convert single values to array, keep arrays as is.
inputs = await callback(inputs);
yield* nextOperation(operation.operations, inputs);
}
/**
*
* @param {PDF|PDF[]} input
* @param {JSON} operation
* @returns {undefined}
*/
async function * oneToN(input, operation, callback) {
if(Array.isArray(input)) {
let output = [];
for (let i = 0; i < input.length; i++) {
output = output.concat(await callback(input[i]));
}
yield* nextOperation(operation.operations, output);
}
else {
input = await callback(input);
yield* nextOperation(operation.operations, input);
}
}
/**
*
* @param {PDF|PDF[]} input
* @param {JSON} operation
* @returns {undefined}
*/
async function * nToN(input, operation, callback) {
if(Array.isArray(input)) {
for (let i = 0; i < input.length; i++) {
await callback(input[i]);
}
yield* nextOperation(operation.operations, input);
}
else {
await callback(input);
yield* nextOperation(operation.operations, input);
}
}
}