GoScrobble/web/node_modules/webpack/lib/Template.js

292 lines
8.0 KiB
JavaScript

/*
MIT License http://www.opensource.org/licenses/mit-license.php
Author Tobias Koppers @sokra
*/
/** @typedef {import("./Module")} Module */
/** @typedef {import("./Chunk")} Chunk */
/** @typedef {import("./ModuleTemplate")} ModuleTemplate */
/** @typedef {import("webpack-sources").ConcatSource} ConcatSource */
const { ConcatSource } = require("webpack-sources");
const HotUpdateChunk = require("./HotUpdateChunk");
const START_LOWERCASE_ALPHABET_CODE = "a".charCodeAt(0);
const START_UPPERCASE_ALPHABET_CODE = "A".charCodeAt(0);
const DELTA_A_TO_Z = "z".charCodeAt(0) - START_LOWERCASE_ALPHABET_CODE + 1;
const FUNCTION_CONTENT_REGEX = /^function\s?\(\)\s?\{\r?\n?|\r?\n?\}$/g;
const INDENT_MULTILINE_REGEX = /^\t/gm;
const LINE_SEPARATOR_REGEX = /\r?\n/g;
const IDENTIFIER_NAME_REPLACE_REGEX = /^([^a-zA-Z$_])/;
const IDENTIFIER_ALPHA_NUMERIC_NAME_REPLACE_REGEX = /[^a-zA-Z0-9$]+/g;
const COMMENT_END_REGEX = /\*\//g;
const PATH_NAME_NORMALIZE_REPLACE_REGEX = /[^a-zA-Z0-9_!§$()=\-^°]+/g;
const MATCH_PADDED_HYPHENS_REPLACE_REGEX = /^-|-$/g;
/** @typedef {import("webpack-sources").Source} Source */
/**
* @typedef {Object} HasId
* @property {number | string} id
*/
/**
* @typedef {function(Module, number): boolean} ModuleFilterPredicate
*/
/**
* @param {HasId} a first id object to be sorted
* @param {HasId} b second id object to be sorted against
* @returns {-1|0|1} the sort value
*/
const stringifyIdSortPredicate = (a, b) => {
const aId = a.id + "";
const bId = b.id + "";
if (aId < bId) return -1;
if (aId > bId) return 1;
return 0;
};
class Template {
/**
*
* @param {Function} fn a runtime function (.runtime.js) "template"
* @returns {string} the updated and normalized function string
*/
static getFunctionContent(fn) {
return fn
.toString()
.replace(FUNCTION_CONTENT_REGEX, "")
.replace(INDENT_MULTILINE_REGEX, "")
.replace(LINE_SEPARATOR_REGEX, "\n");
}
/**
* @param {string} str the string converted to identifier
* @returns {string} created identifier
*/
static toIdentifier(str) {
if (typeof str !== "string") return "";
return str
.replace(IDENTIFIER_NAME_REPLACE_REGEX, "_$1")
.replace(IDENTIFIER_ALPHA_NUMERIC_NAME_REPLACE_REGEX, "_");
}
/**
*
* @param {string} str string to be converted to commented in bundle code
* @returns {string} returns a commented version of string
*/
static toComment(str) {
if (!str) return "";
return `/*! ${str.replace(COMMENT_END_REGEX, "* /")} */`;
}
/**
*
* @param {string} str string to be converted to "normal comment"
* @returns {string} returns a commented version of string
*/
static toNormalComment(str) {
if (!str) return "";
return `/* ${str.replace(COMMENT_END_REGEX, "* /")} */`;
}
/**
* @param {string} str string path to be normalized
* @returns {string} normalized bundle-safe path
*/
static toPath(str) {
if (typeof str !== "string") return "";
return str
.replace(PATH_NAME_NORMALIZE_REPLACE_REGEX, "-")
.replace(MATCH_PADDED_HYPHENS_REPLACE_REGEX, "");
}
// map number to a single character a-z, A-Z or <_ + number> if number is too big
/**
*
* @param {number} n number to convert to ident
* @returns {string} returns single character ident
*/
static numberToIdentifer(n) {
// lower case
if (n < DELTA_A_TO_Z) {
return String.fromCharCode(START_LOWERCASE_ALPHABET_CODE + n);
}
// upper case
if (n < DELTA_A_TO_Z * 2) {
return String.fromCharCode(
START_UPPERCASE_ALPHABET_CODE + n - DELTA_A_TO_Z
);
}
// use multiple letters
return (
Template.numberToIdentifer(n % (2 * DELTA_A_TO_Z)) +
Template.numberToIdentifer(Math.floor(n / (2 * DELTA_A_TO_Z)))
);
}
/**
*
* @param {string | string[]} s string to convert to identity
* @returns {string} converted identity
*/
static indent(s) {
if (Array.isArray(s)) {
return s.map(Template.indent).join("\n");
} else {
const str = s.trimRight();
if (!str) return "";
const ind = str[0] === "\n" ? "" : "\t";
return ind + str.replace(/\n([^\n])/g, "\n\t$1");
}
}
/**
*
* @param {string|string[]} s string to create prefix for
* @param {string} prefix prefix to compose
* @returns {string} returns new prefix string
*/
static prefix(s, prefix) {
const str = Template.asString(s).trim();
if (!str) return "";
const ind = str[0] === "\n" ? "" : prefix;
return ind + str.replace(/\n([^\n])/g, "\n" + prefix + "$1");
}
/**
*
* @param {string|string[]} str string or string collection
* @returns {string} returns a single string from array
*/
static asString(str) {
if (Array.isArray(str)) {
return str.join("\n");
}
return str;
}
/**
* @typedef {Object} WithId
* @property {string|number} id
*/
/**
* @param {WithId[]} modules a collection of modules to get array bounds for
* @returns {[number, number] | false} returns the upper and lower array bounds
* or false if not every module has a number based id
*/
static getModulesArrayBounds(modules) {
let maxId = -Infinity;
let minId = Infinity;
for (const module of modules) {
if (typeof module.id !== "number") return false;
if (maxId < module.id) maxId = /** @type {number} */ (module.id);
if (minId > module.id) minId = /** @type {number} */ (module.id);
}
if (minId < 16 + ("" + minId).length) {
// add minId x ',' instead of 'Array(minId).concat(…)'
minId = 0;
}
const objectOverhead = modules
.map(module => (module.id + "").length + 2)
.reduce((a, b) => a + b, -1);
const arrayOverhead =
minId === 0 ? maxId : 16 + ("" + minId).length + maxId;
return arrayOverhead < objectOverhead ? [minId, maxId] : false;
}
/**
* @param {Chunk} chunk chunk whose modules will be rendered
* @param {ModuleFilterPredicate} filterFn function used to filter modules from chunk to render
* @param {ModuleTemplate} moduleTemplate ModuleTemplate instance used to render modules
* @param {TODO | TODO[]} dependencyTemplates templates needed for each module to render dependencies
* @param {string=} prefix applying prefix strings
* @returns {ConcatSource} rendered chunk modules in a Source object
*/
static renderChunkModules(
chunk,
filterFn,
moduleTemplate,
dependencyTemplates,
prefix = ""
) {
const source = new ConcatSource();
const modules = chunk.getModules().filter(filterFn);
let removedModules;
if (chunk instanceof HotUpdateChunk) {
removedModules = chunk.removedModules;
}
if (
modules.length === 0 &&
(!removedModules || removedModules.length === 0)
) {
source.add("[]");
return source;
}
/** @type {{id: string|number, source: Source|string}[]} */
const allModules = modules.map(module => {
return {
id: module.id,
source: moduleTemplate.render(module, dependencyTemplates, {
chunk
})
};
});
if (removedModules && removedModules.length > 0) {
for (const id of removedModules) {
allModules.push({
id,
source: "false"
});
}
}
const bounds = Template.getModulesArrayBounds(allModules);
if (bounds) {
// Render a spare array
const minId = bounds[0];
const maxId = bounds[1];
if (minId !== 0) {
source.add(`Array(${minId}).concat(`);
}
source.add("[\n");
/** @type {Map<string|number, {id: string|number, source: Source|string}>} */
const modules = new Map();
for (const module of allModules) {
modules.set(module.id, module);
}
for (let idx = minId; idx <= maxId; idx++) {
const module = modules.get(idx);
if (idx !== minId) {
source.add(",\n");
}
source.add(`/* ${idx} */`);
if (module) {
source.add("\n");
source.add(module.source);
}
}
source.add("\n" + prefix + "]");
if (minId !== 0) {
source.add(")");
}
} else {
// Render an object
source.add("{\n");
allModules.sort(stringifyIdSortPredicate).forEach((module, idx) => {
if (idx !== 0) {
source.add(",\n");
}
source.add(`\n/***/ ${JSON.stringify(module.id)}:\n`);
source.add(module.source);
});
source.add(`\n\n${prefix}}`);
}
return source;
}
}
module.exports = Template;