Interesting that the same problem (high churn) leads to two polarly opposite conclusions - i.e. the article suggests going with many specialized libraries, whereas Ember is on the camp of incorporating the latest and greatest into a single framework when it makes sense.
My opinion is that some types of features are better off bundled in a ready-to-go package. For example data binding and ajax almost always go together and just plopping a CDN link to Angular is a lot easier than installing NPM, setting up a file watcher and who knows what else to get React's JSX running (and you still need to go decide what you're going to use for AJAX).
Other less "core" types of features are better off as standalone libraries because their depth of usage can vary from not being used at all to requiring very domain-specific features (e.g. accounting.js is a good example of this). I don't agree with the author's premise that micro-libraries are easier to swap. I'd be very doubtful if anyone were to claim that it's trivial to convert a React codebase to Knockout, for example. You'd likely be hard-pressed to swap even things like moment.js because you probably chose the library based on it being better than its alternatives to begin with.
But having these types of features be standalone libraries does provide some benefits, e.g. being able to opt-out of a breaking release in a more granular fashion, trial-and-error-oriented development, etc.
My opinion is that some types of features are better off bundled in a ready-to-go package. For example data binding and ajax almost always go together and just plopping a CDN link to Angular is a lot easier than installing NPM, setting up a file watcher and who knows what else to get React's JSX running (and you still need to go decide what you're going to use for AJAX).
Other less "core" types of features are better off as standalone libraries because their depth of usage can vary from not being used at all to requiring very domain-specific features (e.g. accounting.js is a good example of this). I don't agree with the author's premise that micro-libraries are easier to swap. I'd be very doubtful if anyone were to claim that it's trivial to convert a React codebase to Knockout, for example. You'd likely be hard-pressed to swap even things like moment.js because you probably chose the library based on it being better than its alternatives to begin with.
But having these types of features be standalone libraries does provide some benefits, e.g. being able to opt-out of a breaking release in a more granular fashion, trial-and-error-oriented development, etc.