Skipped 1 file(s) >100MB (GitHub limit): source/__polyfill/04xxx/module_4821.js (118.4 MB)
120 lines
3.6 KiB
JavaScript
120 lines
3.6 KiB
JavaScript
// Module ID: 5266
|
|
// Function ID: 5267
|
|
// Dependencies: []
|
|
// Exports: findMatroskaDocTypeElements, getFileChunk, isAvifStringIncluded, isFileContaineJfiforExifHeader, isFlvStringIncluded, isHeicSignatureIncluded, isftypStringIncluded
|
|
|
|
// Module 5266
|
|
function fetchFromObject(FileTypes, arr) {
|
|
const index = arr.indexOf(".");
|
|
if (index > -1) {
|
|
let tmp2 = fetchFromObject(FileTypes[arr.slice(arr, 0, index)], arr.slice(index + 1));
|
|
} else {
|
|
tmp2 = FileTypes[arr];
|
|
}
|
|
return tmp2;
|
|
}
|
|
|
|
export const getFileChunk = function getFileChunk(fileChunk, chunkSize) {
|
|
let num = chunkSize;
|
|
if (chunkSize === undefined) {
|
|
num = 32;
|
|
}
|
|
let uint8Array = fileChunk;
|
|
if (fileChunk instanceof ArrayBuffer) {
|
|
const _Uint8Array = Uint8Array;
|
|
uint8Array = new Uint8Array(fileChunk);
|
|
}
|
|
if (!Array.isArray(fileChunk)) {
|
|
const _ArrayBuffer = ArrayBuffer;
|
|
if (!(fileChunk instanceof ArrayBuffer)) {
|
|
const _Uint8Array2 = Uint8Array;
|
|
if (!(fileChunk instanceof Uint8Array)) {
|
|
const _TypeError = TypeError;
|
|
const _HermesInternal = HermesInternal;
|
|
const typeError = new TypeError("Expected the `file` argument to be of type `Array<number>`, `Uint8Array`, or `ArrayBuffer`, got `" + typeof fileChunk + "`");
|
|
throw typeError;
|
|
}
|
|
}
|
|
}
|
|
const arr = Array.from(uint8Array.slice(0, num));
|
|
if (arr.every((item) => {
|
|
let tmp = typeof item === "number";
|
|
if (typeof item === "number") {
|
|
const _isNaN = isNaN;
|
|
tmp = !isNaN(item);
|
|
}
|
|
return tmp;
|
|
})) {
|
|
return arr;
|
|
} else {
|
|
const _TypeError2 = TypeError;
|
|
const typeError1 = new TypeError("File content contains illegal values");
|
|
throw typeError1;
|
|
}
|
|
};
|
|
export { fetchFromObject };
|
|
export const findMatroskaDocTypeElements = function findMatroskaDocTypeElements(fileChunk) {
|
|
const mapped = fileChunk.map((item) => String.fromCharCode(item));
|
|
const joined = mapped.join("");
|
|
let str = "webm";
|
|
if (!joined.includes("webm")) {
|
|
let str3;
|
|
if (joined.includes("matroska")) {
|
|
str3 = "mkv";
|
|
}
|
|
str = str3;
|
|
}
|
|
return str;
|
|
};
|
|
export const isftypStringIncluded = function isftypStringIncluded(fileChunk) {
|
|
const items = [102, 116, 121, 112];
|
|
let num = 0;
|
|
if (0 < fileChunk.length - items.length) {
|
|
while (true) {
|
|
let num2 = 0;
|
|
let flag = true;
|
|
if (0 < items.length) {
|
|
flag = false;
|
|
while (fileChunk[num + num2] === items[num2]) {
|
|
let sum = num2 + 1;
|
|
num2 = sum;
|
|
flag = true;
|
|
if (sum >= items.length) {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
if (flag) {
|
|
break;
|
|
} else {
|
|
num = num + 1;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
return false;
|
|
};
|
|
export const isFlvStringIncluded = function isFlvStringIncluded(fileChunk) {
|
|
const substr = fileChunk.slice(0, 3);
|
|
const decoder = new TextDecoder();
|
|
const uint8Array = new Uint8Array(substr);
|
|
return decoder.decode(uint8Array).includes("FLV");
|
|
};
|
|
export const isFileContaineJfiforExifHeader = function isFileContaineJfiforExifHeader(arg0) {
|
|
let tmp2 = 224 === tmp;
|
|
if (!tmp2) {
|
|
tmp2 = 225 === tmp;
|
|
}
|
|
return tmp2;
|
|
};
|
|
export const isAvifStringIncluded = function isAvifStringIncluded(fileChunk) {
|
|
const substr = fileChunk.slice(4, 12);
|
|
const mapped = substr.map((item) => String.fromCharCode(item));
|
|
return "ftypavif" === mapped.join("");
|
|
};
|
|
export const isHeicSignatureIncluded = function isHeicSignatureIncluded(fileChunk) {
|
|
const mapped = fileChunk.map((item) => String.fromCharCode(item));
|
|
closure_0 = mapped.join("");
|
|
const items = ["ftypheic", "ftyphevc", "ftypmif1", "ftypmsf1"];
|
|
return items.some((item) => closure_0.includes(item));
|
|
};
|