In 2014 I still think my point holds:
IMHO, this discussion got blown out of proportion quite a bit. Quoting the aforementioned blog post:
Most JavaScript utility libraries, such as Underscore, Valentine, and
wu, rely on the “native-first dual approach.” This approach prefers
native implementations, falling back to vanilla JavaScript only if the
native equivalent is not supported. But jsPerf revealed an interesting
trend: the most efficient way to iterate over an array or array-like
collection is to avoid the native implementations entirely, opting for
simple loops instead.
As if "simple loops" and "vanilla Javascript" are more native than Array or Object method implementations. Jeez ...
It certainly would be nice to have a single source of truth, but there isn't. Even if you've been told otherwise, there is no Vanilla God, my dear. I'm sorry. The only assumption that really holds is that we are all writing JavaScript code that aims at performing well in all major browsers, knowing that all of them have different implementations of the same things. It's a bitch to cope with, to put it mildly. But that's the premise, whether you like it or not.
Maybe all of you are working on large scale projects that need twitterish performance so that you really see the difference between 850,000 (Underscore.js) vs. 2,500,000 (Lodash) iterations over a list per second right now!
I for one am not. I mean, I worked on projects where I had to address performance issues, but they were never solved or caused by neither Underscore.js nor Lodash. And unless I get hold of the real differences in implementation and performance (we're talking C++ right now) of, let’s say, a loop over an iterable (object or array, sparse or not!), I rather don't get bothered with any claims based on the results of a benchmark platform that is already opinionated.
It only needs one single update of, let’s say, Rhino to set its Array method implementations on fire in a fashion that not a single "medieval loop methods perform better and forever and whatnot" priest can argue his/her way around the simple fact that all of a sudden array methods in Firefox are much faster than his/her opinionated brainfuck. Man, you just can't cheat your runtime environment by cheating your runtime environment! Think about that when promoting ...
your utility belt
... next time.
So to keep it relevant:
- Use Underscore.js if you're into convenience without sacrificing native'ish.
- Use Lodash if you're into convenience and like its extended feature catalogue (deep copy, etc.) and if you're in desperate need of instant performance and most importantly don't mind settling for an alternative as soon as native API's outshine opinionated workarounds. Which is going to happen soon. Period.
- There's even a third solution. DIY! Know your environments. Know about inconsistencies. Read their (John-David's and Jeremy's) code. Don't use this or that without being able to explain why a consistency/compatibility layer is really needed and enhances your workflow or improves the performance of your application. It is very likely that your requirements are satisfied with a simple polyfill that you're perfectly able to write yourself. Both libraries are just plain vanilla with a little bit of sugar. They both just fight over who's serving the sweetest pie. But believe me, in the end both are only cooking with water. There's no Vanilla God so there can't be no Vanilla pope, right?
Choose whatever approach fits your needs the most. As usual. I'd prefer fallbacks on actual implementations over opinionated runtime cheats anytime, but even that seems to be a matter of taste nowadays. Stick to quality resources like http://developer.mozilla.com and http://caniuse.com and you'll be just fine.
