Hacker News Reader

Comment by lightswitch05 | Hacker News Reader

I’ve written a library with zero production dependencies. Of course I have Jest as a development dependency which pulls in all sorts of stuff. It would be difficult to make a zero-dependency library. As for not having production libraries, this was my experience:

1. Using the https module directly was more work than I expected, especially with error handling. This made me really look forward to the new Fetch API coming out.

2. No CLI parser. Its not like parsing args is a LOT of work - but its also something that is already solved and having to write support for that directly was a bummer

3. No logging library. This one was pretty easy. Create a little class with logging levels. Again this is something that is very common that would have been nice to use a package for.

View on HN

What you describe is exactly why people use dependencies. You just decided to trade your time for the noble act of having “no production dependencies”, while one of the 275 modules installed by Jest (real number) stole your production secrets anyway.

As for point 2, Node 18.1 I think just introduced a native argument parser.

Why would the test dependencies have access to production secrets? They only get installed while developing

They still get run on a developer’s machine most of the time and are at least installed there where they can run arbitrary code on install. And there are juicy secrets beyond just production server secrets sitting on your laptop.

You don't need Jest. You can use buld-in test runner since Node 19.0 (https://nodejs.org/api/test.html)

1/ You do not need to use https / https module directly. You can use Fetch API since Node 17.5 ( https://nodejs.org/api/globals.html#fetch ).

2/ You do not need an external CLI parser. You can use build-in CLI parser since Node v18.3 ( https://nodejs.org/api/util.html#utilparseargsconfig )

> You don't need Jest. You can use buld-in test runner since Node 19.0 (https://nodejs.org/api/test.html)

v19 was released 5 days ago. Barely had time to register this was a thing, let alone migrating our 3500+ unit test+ 300 integration test suite lol

> 1/ You do not need to use https / https module directly. You can use Fetch API since Node 17.5 ( https://nodejs.org/api/globals.html#fetch ). > 2/ You do not need an external CLI parser. You can use build-in CLI parser since Node v18.3 ( https://nodejs.org/api/util.html#utilparseargsconfig )

Lots of production environments tend to sit on latest LTS. v18 _just_ hit LTS, up to last month it was v16. Those weren't really an option for all these projects until then.

Yes, my library absolutely has to support the LTS version of Node, and I run the tests against all supported versions to ensure compatibility. So, one day I can use the nice things people are mentioning, but it will be years from now.

I tried making a tiny test runner that printed nicely to the console. It’s very basic, but I was surprised at how little code it took: https://github.com/jcpst/neon-johnston/blob/master/test/inde...

> I've written a library with zero production dependencies. Of course I have Jest as a development dependency which pulls in all sorts of stuff. It would be difficult to make a zero-dependency library.

That really depends on what your expectations and goals are. You mention Jest but as it's a test dependency then security is not a major concern, and thus it's ok to just vend the source code into your project trees.

> No logging library. This one was pretty easy. Create a little class with logging levels.

Do you print to /dev/stdlog using the correct syslog format or use journald and its format?

Because if it's not one of those, the "levels" are useless.

No, the levels are not useless, far from it. Even if you just print to stdout, prefixing each line with a (maybe colored?) level info is very useful. Also, if later you decide to use xyz format, it is trivial to implement this. Changing all logging calls? Not so much.

Just use a real logging daemon and it will be the colours by itself.

It will also let you query for which level you want.

> Changing all logging calls? Not so much.

Uh? Make a good library using the same interface.

To answer your question, the levels are important for both regular usage and debugging when something goes wrong. I create a number of debug/info/warning/error log messages throughout the tool and allow the desired level of logging to be set as a CLI argument - or even run in silent mode if desired.

Reply on news.ycombinator.com