Yes, that's `npm ci`.
With a fresh `npm install` you get the latest packages that satisfy dependencies in the ranges spec'd in package.json. That's basically a crap shoot because even if you specifically pin the versions of all your dependencies, chances are that at least some of your dependencies did not have the good sense to do the same for their dependencies. This is the path to hell.
With a fresh `npm ci` on the other hand, you get back exactly to the known state specified in package-lock.json, where everything is pinned, all the way down. This is the path to happiness.
> That's basically a crap shoot because even if you specifically pin the versions of all your dependencies, chances are that at least some of your dependencies did not have the good sense to do the same for their dependencies.
As I mentioned in my first comment, I don't really want my dependencies to pin their dependencies. I want them to specify a range they work with to minimize the number of redundant copies of common transitive dependencies.
That phrase is frequently way more load-bearing than it should be. Is pnpm's default behavior the correct one? (yarn user here, but open to switching)
No tool can prevent all footguns, but standard / idiomatic / out-of-the-box pnpm usage beats even expert use of npm or yarn. It's different, and better.
2- I want every install of my project- be it on a dev machine or deploy machine- to have the exact same versions of dependencies, including transitive ones. I don't want to deal with bugs caused by surprise version changes
3- if I upgrade a dependency or remove it, I want the transitive dependencies managed automatically. I don't want orphaned transitive dependencies. In fact, I don't even want to think about them at all other than know that they work and aren't adding bloat or security risks.
You need a package manager with more than a passing thought for handling lock files. For the longest time, npm wasn't it. I'd argue that it still isn't, because "npm install" should NOT be the command used for both "set up a project for the first time" and "add a new package". In the first case, I want a reproducible, deterministic result of a known state. In the second case, I want to modify the dependency graph, producing a new state.