Simple, smart hot reloading - now for the backend.
DotHot allows you to edit your running Node.js application without needing to restart it as often (or at all.)
DotHot watches require'd files for changes (using chokidar).
When you edit a module, DotHot deletes it from require.cache.
All existing references to things from that module are preserved as-is.
However, if you require the module anew, a curious thing happens:
require actually re-evaluates your module and gives you the updated version!
Now if, instead of this:
var {doSomething} = require('module')
myFunction () {
doSomething()
}you do this:
myFunction () {
require('module').doSomething()
}You can get away with restarting Node a lot less often!
Needless to say, it makes little sense to run this in production.
Furthermore, it is up to you to avoid memory leaks from global objects that fail to be dereferenced upon deleting the module object (though a restart of your program should clean those up.)
npm i dothot
node -r dothot your_main_script.js
The env var NODE_HOT_OUT can be stdout, stderr or a path to a file;
if it is set, debugging info is printed to that destination.
The main competitor in the space of preventing restarts to the Node.js process
and saving Node developers valuable time is nodemon (kudos @remy!). Here's
a quick comparison/list of reasons why you may prefer one over the other.
- Nodemon has more features, and is coding style agnostic. In fact, since it works on the filesystem level, it supports reloading programs in other languages besides JS! On the other hand, DotHot's design reminds you to keep state under control and write in a concise, functional style.
- DotHot does exactly one thing, and is ~150LOC, pulling in another ~4000 LOC
of dependencies in 20
node_modules; for comparison, Nodemon is around ~2000LOC (which is pretty sweet) but depends on 102 NPM modules weighing in at another ~17000 LOC. - Restarting your whole application every time a file changes is slower, and loses any in-memory state. This is especially noticeable if you're doing any heavy lifting at startup, perhaps prompting you to reach for an external datastore and a distributed paradigm where a more self-contained approach would've been cleaner.
Remember, software isn't going to solve itself on its own but is known to do the opposite.