Introduction
For beginners, webpack building performance is sometimes annoying. However, this will not stop webpack from becoming the best web building tool in the world. Once you master the following skills, you can soon boost the building performance and save your team a lot of time.
For one thing before kickoff, webpack.js.org is now the official site for webpack. I browsed the documentation and soon found out that this one is much better than the old doc. Though it is not yet finished, it deserves half an hour reading.
Overall config for webpack
Here is the overall configuration for webpack.
entry
is for setting up source code file.output
is for setting up output cdn, destination and filename.loader
is the config for loaders to compile files from certain format to the format which can be recognized by the browser.plugins
: via plugin, developer can utilize more webpack apis.resolve
is the guide for webpack to search files before real compilation
Architecture Design For webpack Config
As gulp
fans, many developers highly rely on the work flow of gulp. So they tend to use gulp as the core to control the building flow, and only use webpack to pack js files. Based on this habbit, they usually use webpack-stream which I think, is a bad idea at most cases.
Since they usually use gulp.watch
at developing mode, once they modify a file, the whole gulp building process will rerun which takes time. Of course, you can write some building logic to prevent some building process from being invoked, however, it is far from good.
I used to test this idea using a big project with, it takes 13 seconds for incremental build under developing mode.
Based on my experience, webpack is more than just a packing tool, as the community grows and the number of plugins and loaders flourishes, it becomes a solid competitor agains gulp
and grunt
.
I intend to use webpack as the core of the whold building tool for the project. gulp
or grunt
is used only for some tasks which webpack is still not good at, like multiple sprites. For flow control, you can use package.json
to manage.
&
is for serial&&
is for parallelexport
is used in Mac or Linux for injecting environment variableset
is used in Windows for injecting environment variable
Comprehensive Guide of webpack Performance for Big-scale Project
I use a react
project of 10 js files as an example. I run the test in a machine with 2 CPU cores and 8G ram. For a very basic configuration (without file uglification), it takes 26seconds.
Let me summarize the tips first.
- Externalize big libraries
- Pre-build files
- Reduce searching time
- Cache
- Parallel working
Externalize big libraries
Put the big libraries like react
, zepto
in script tag directly, and use external
config to tell webpack, it is fine to use the external link as the source, don't build it. It saves 8 seconds.
Pre-build files
In this case, we are trying to pre-build some files before we really start the main building flow.
- DllPlugin and DllReferencePlugin
First, we create a new config file,webpack.dll.js
in which we use DllPlugin. Please see the following pic, we useDllPlugin
to pre-buildreact
,react-dom
,redux
,react-redux
,redux-thunk
andloadash.merge
as a new vendor calledlib.js
. At the same time, a dependencies file./tool/manifest.json
is created for next-step usage.
In main webpack config, we now use DllReferencePlugn
to reference the dependencies.
- noParse
Once you have some files which are in ES5 format, you don't have to parse it.
- PrefetchPlugin
This plugin maybe strange to most developers. I do not completely master it either. For this plugin, it cost extra analysis since it varies from project to project.
Firstly, stats-webpack-plugin is used for generating a work flow file, stats.json
.
Secondly, enter this site webpack.github.io/analyse/ and upload stats.json
. If you concern about code security, you can checkout this project to your local machine and do the analysis work in local environment.
Thirdly, click hint
and locate long module build chains
part. If this part exists, you can prefetch
part of the long building chain like the following:
Due to conflict of external
and dll
, I don't use external
config this time. However it also saves much time. Reduce from 26 seconds to 16.8 seconds.
Reduce searching time
Before webpack compiles files, it needs to locate each file. Searching files is also a time-consuming job. Please cut it when necessary.
- loader exclude
Some times, loader can skip some folders to avoid unnecessary searching. For example,babel-loader
can skipnode_modules
.
resolve.alias
You can also directly tell webpack where the library is viaresolve.alias
.
resolve.unsafeCache
If the libraries are stable, you can useresolve.unsafeCache
to save the file location in webpack so that next time, webpack does not need to search again. Set it astrue
. Once the a library changes its location, please disable it and make webpack search agagin.Be careful when you use
resolve.root
,resolve.modulesDirectory
andresolve.fallback
When you encounters these three config, please be careful. If you set the wrong searching path or if you set to many, it lower down the compilation time.
After these optimization, it further reduce the building time from 18 seconds to 15 seconds.
Cache
We can make a big progress using cache. For example, babel-loader
has a cache option. As JavaScript takes up the majority of compilation time when we enable it, it nearly reduces half of the compilation time, from 15 seconds to 8 seconds.
Parallel Working
In this section, I introduce a famous parallel packing tool, happypack for webpack.
For js, please just directly copy the options from loader config.
For css, you can use it only in developing mode since it does not support extract-text-webpack-plugin
.
3 more seconds are decreased.
Conclusion
So far, we have reduce the total compilation time from 26 seconds to 5 seconds, a remarkable result.
What about the compilation time after uglification?
If we use the basic config with webpack.optimize.UglifyJsPlugin
, it takes 66 seconds to finish the whole process. However, if we use webpack-uglify-parallel
, another parallel uglification tool and combine suitable optimazation tips introduced above, it takes only 19.5 seconds.