322 lines
9.8 KiB
JavaScript
322 lines
9.8 KiB
JavaScript
/*
|
|
MIT License http://www.opensource.org/licenses/mit-license.php
|
|
Author Tobias Koppers @sokra
|
|
*/
|
|
|
|
"use strict";
|
|
|
|
const { validate } = require("schema-utils");
|
|
const schema = require("../../schemas/plugins/optimize/AggressiveSplittingPlugin.json");
|
|
const { STAGE_ADVANCED } = require("../OptimizationStages");
|
|
const { intersect } = require("../util/SetHelpers");
|
|
const {
|
|
compareModulesByIdentifier,
|
|
compareChunks
|
|
} = require("../util/comparators");
|
|
const identifierUtils = require("../util/identifier");
|
|
|
|
/** @typedef {import("../../declarations/plugins/optimize/AggressiveSplittingPlugin").AggressiveSplittingPluginOptions} AggressiveSplittingPluginOptions */
|
|
/** @typedef {import("../Chunk")} Chunk */
|
|
/** @typedef {import("../ChunkGraph")} ChunkGraph */
|
|
/** @typedef {import("../Compiler")} Compiler */
|
|
/** @typedef {import("../Module")} Module */
|
|
|
|
const moveModuleBetween = (chunkGraph, oldChunk, newChunk) => {
|
|
return module => {
|
|
chunkGraph.disconnectChunkAndModule(oldChunk, module);
|
|
chunkGraph.connectChunkAndModule(newChunk, module);
|
|
};
|
|
};
|
|
|
|
/**
|
|
* @param {ChunkGraph} chunkGraph the chunk graph
|
|
* @param {Chunk} chunk the chunk
|
|
* @returns {function(Module): boolean} filter for entry module
|
|
*/
|
|
const isNotAEntryModule = (chunkGraph, chunk) => {
|
|
return module => {
|
|
return !chunkGraph.isEntryModuleInChunk(module, chunk);
|
|
};
|
|
};
|
|
|
|
/** @type {WeakSet<Chunk>} */
|
|
const recordedChunks = new WeakSet();
|
|
|
|
class AggressiveSplittingPlugin {
|
|
/**
|
|
* @param {AggressiveSplittingPluginOptions=} options options object
|
|
*/
|
|
constructor(options = {}) {
|
|
validate(schema, options, {
|
|
name: "Aggressive Splitting Plugin",
|
|
baseDataPath: "options"
|
|
});
|
|
|
|
this.options = options;
|
|
if (typeof this.options.minSize !== "number") {
|
|
this.options.minSize = 30 * 1024;
|
|
}
|
|
if (typeof this.options.maxSize !== "number") {
|
|
this.options.maxSize = 50 * 1024;
|
|
}
|
|
if (typeof this.options.chunkOverhead !== "number") {
|
|
this.options.chunkOverhead = 0;
|
|
}
|
|
if (typeof this.options.entryChunkMultiplicator !== "number") {
|
|
this.options.entryChunkMultiplicator = 1;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* @param {Chunk} chunk the chunk to test
|
|
* @returns {boolean} true if the chunk was recorded
|
|
*/
|
|
static wasChunkRecorded(chunk) {
|
|
return recordedChunks.has(chunk);
|
|
}
|
|
|
|
/**
|
|
* Apply the plugin
|
|
* @param {Compiler} compiler the compiler instance
|
|
* @returns {void}
|
|
*/
|
|
apply(compiler) {
|
|
compiler.hooks.thisCompilation.tap(
|
|
"AggressiveSplittingPlugin",
|
|
compilation => {
|
|
let needAdditionalSeal = false;
|
|
let newSplits;
|
|
let fromAggressiveSplittingSet;
|
|
let chunkSplitDataMap;
|
|
compilation.hooks.optimize.tap("AggressiveSplittingPlugin", () => {
|
|
newSplits = [];
|
|
fromAggressiveSplittingSet = new Set();
|
|
chunkSplitDataMap = new Map();
|
|
});
|
|
compilation.hooks.optimizeChunks.tap(
|
|
{
|
|
name: "AggressiveSplittingPlugin",
|
|
stage: STAGE_ADVANCED
|
|
},
|
|
chunks => {
|
|
const chunkGraph = compilation.chunkGraph;
|
|
// Precompute stuff
|
|
const nameToModuleMap = new Map();
|
|
const moduleToNameMap = new Map();
|
|
const makePathsRelative = identifierUtils.makePathsRelative.bindContextCache(
|
|
compiler.context,
|
|
compiler.root
|
|
);
|
|
for (const m of compilation.modules) {
|
|
const name = makePathsRelative(m.identifier());
|
|
nameToModuleMap.set(name, m);
|
|
moduleToNameMap.set(m, name);
|
|
}
|
|
|
|
// Check used chunk ids
|
|
const usedIds = new Set();
|
|
for (const chunk of chunks) {
|
|
usedIds.add(chunk.id);
|
|
}
|
|
|
|
const recordedSplits =
|
|
(compilation.records && compilation.records.aggressiveSplits) ||
|
|
[];
|
|
const usedSplits = newSplits
|
|
? recordedSplits.concat(newSplits)
|
|
: recordedSplits;
|
|
|
|
const minSize = this.options.minSize;
|
|
const maxSize = this.options.maxSize;
|
|
|
|
const applySplit = splitData => {
|
|
// Cannot split if id is already taken
|
|
if (splitData.id !== undefined && usedIds.has(splitData.id)) {
|
|
return false;
|
|
}
|
|
|
|
// Get module objects from names
|
|
const selectedModules = splitData.modules.map(name =>
|
|
nameToModuleMap.get(name)
|
|
);
|
|
|
|
// Does the modules exist at all?
|
|
if (!selectedModules.every(Boolean)) return false;
|
|
|
|
// Check if size matches (faster than waiting for hash)
|
|
let size = 0;
|
|
for (const m of selectedModules) size += m.size();
|
|
if (size !== splitData.size) return false;
|
|
|
|
// get chunks with all modules
|
|
const selectedChunks = intersect(
|
|
selectedModules.map(
|
|
m => new Set(chunkGraph.getModuleChunksIterable(m))
|
|
)
|
|
);
|
|
|
|
// No relevant chunks found
|
|
if (selectedChunks.size === 0) return false;
|
|
|
|
// The found chunk is already the split or similar
|
|
if (
|
|
selectedChunks.size === 1 &&
|
|
chunkGraph.getNumberOfChunkModules(
|
|
Array.from(selectedChunks)[0]
|
|
) === selectedModules.length
|
|
) {
|
|
const chunk = Array.from(selectedChunks)[0];
|
|
if (fromAggressiveSplittingSet.has(chunk)) return false;
|
|
fromAggressiveSplittingSet.add(chunk);
|
|
chunkSplitDataMap.set(chunk, splitData);
|
|
return true;
|
|
}
|
|
|
|
// split the chunk into two parts
|
|
const newChunk = compilation.addChunk();
|
|
newChunk.chunkReason = "aggressive splitted";
|
|
for (const chunk of selectedChunks) {
|
|
selectedModules.forEach(
|
|
moveModuleBetween(chunkGraph, chunk, newChunk)
|
|
);
|
|
chunk.split(newChunk);
|
|
chunk.name = null;
|
|
}
|
|
fromAggressiveSplittingSet.add(newChunk);
|
|
chunkSplitDataMap.set(newChunk, splitData);
|
|
|
|
if (splitData.id !== null && splitData.id !== undefined) {
|
|
newChunk.id = splitData.id;
|
|
newChunk.ids = [splitData.id];
|
|
}
|
|
return true;
|
|
};
|
|
|
|
// try to restore to recorded splitting
|
|
let changed = false;
|
|
for (let j = 0; j < usedSplits.length; j++) {
|
|
const splitData = usedSplits[j];
|
|
if (applySplit(splitData)) changed = true;
|
|
}
|
|
|
|
// for any chunk which isn't splitted yet, split it and create a new entry
|
|
// start with the biggest chunk
|
|
const cmpFn = compareChunks(chunkGraph);
|
|
const sortedChunks = Array.from(chunks).sort((a, b) => {
|
|
const diff1 =
|
|
chunkGraph.getChunkModulesSize(b) -
|
|
chunkGraph.getChunkModulesSize(a);
|
|
if (diff1) return diff1;
|
|
const diff2 =
|
|
chunkGraph.getNumberOfChunkModules(a) -
|
|
chunkGraph.getNumberOfChunkModules(b);
|
|
if (diff2) return diff2;
|
|
return cmpFn(a, b);
|
|
});
|
|
for (const chunk of sortedChunks) {
|
|
if (fromAggressiveSplittingSet.has(chunk)) continue;
|
|
const size = chunkGraph.getChunkModulesSize(chunk);
|
|
if (
|
|
size > maxSize &&
|
|
chunkGraph.getNumberOfChunkModules(chunk) > 1
|
|
) {
|
|
const modules = chunkGraph
|
|
.getOrderedChunkModules(chunk, compareModulesByIdentifier)
|
|
.filter(isNotAEntryModule(chunkGraph, chunk));
|
|
const selectedModules = [];
|
|
let selectedModulesSize = 0;
|
|
for (let k = 0; k < modules.length; k++) {
|
|
const module = modules[k];
|
|
const newSize = selectedModulesSize + module.size();
|
|
if (newSize > maxSize && selectedModulesSize >= minSize) {
|
|
break;
|
|
}
|
|
selectedModulesSize = newSize;
|
|
selectedModules.push(module);
|
|
}
|
|
if (selectedModules.length === 0) continue;
|
|
const splitData = {
|
|
modules: selectedModules
|
|
.map(m => moduleToNameMap.get(m))
|
|
.sort(),
|
|
size: selectedModulesSize
|
|
};
|
|
|
|
if (applySplit(splitData)) {
|
|
newSplits = (newSplits || []).concat(splitData);
|
|
changed = true;
|
|
}
|
|
}
|
|
}
|
|
if (changed) return true;
|
|
}
|
|
);
|
|
compilation.hooks.recordHash.tap(
|
|
"AggressiveSplittingPlugin",
|
|
records => {
|
|
// 4. save made splittings to records
|
|
const allSplits = new Set();
|
|
const invalidSplits = new Set();
|
|
|
|
// Check if some splittings are invalid
|
|
// We remove invalid splittings and try again
|
|
for (const chunk of compilation.chunks) {
|
|
const splitData = chunkSplitDataMap.get(chunk);
|
|
if (splitData !== undefined) {
|
|
if (splitData.hash && chunk.hash !== splitData.hash) {
|
|
// Split was successful, but hash doesn't equal
|
|
// We can throw away the split since it's useless now
|
|
invalidSplits.add(splitData);
|
|
}
|
|
}
|
|
}
|
|
|
|
if (invalidSplits.size > 0) {
|
|
records.aggressiveSplits = records.aggressiveSplits.filter(
|
|
splitData => !invalidSplits.has(splitData)
|
|
);
|
|
needAdditionalSeal = true;
|
|
} else {
|
|
// set hash and id values on all (new) splittings
|
|
for (const chunk of compilation.chunks) {
|
|
const splitData = chunkSplitDataMap.get(chunk);
|
|
if (splitData !== undefined) {
|
|
splitData.hash = chunk.hash;
|
|
splitData.id = chunk.id;
|
|
allSplits.add(splitData);
|
|
// set flag for stats
|
|
recordedChunks.add(chunk);
|
|
}
|
|
}
|
|
|
|
// Also add all unused historical splits (after the used ones)
|
|
// They can still be used in some future compilation
|
|
const recordedSplits =
|
|
compilation.records && compilation.records.aggressiveSplits;
|
|
if (recordedSplits) {
|
|
for (const splitData of recordedSplits) {
|
|
if (!invalidSplits.has(splitData)) allSplits.add(splitData);
|
|
}
|
|
}
|
|
|
|
// record all splits
|
|
records.aggressiveSplits = Array.from(allSplits);
|
|
|
|
needAdditionalSeal = false;
|
|
}
|
|
}
|
|
);
|
|
compilation.hooks.needAdditionalSeal.tap(
|
|
"AggressiveSplittingPlugin",
|
|
() => {
|
|
if (needAdditionalSeal) {
|
|
needAdditionalSeal = false;
|
|
return true;
|
|
}
|
|
}
|
|
);
|
|
}
|
|
);
|
|
}
|
|
}
|
|
module.exports = AggressiveSplittingPlugin;
|