Hacker News Reader

Comment by davidw | Hacker News Reader

I think you're replying to the wrong person - they just asked a question about when Erlang might be a good fit.

> Just call into a C lib from Erlang.

That's easier said than done. Certainly possible, but it's not quite as easy as just linking and calling.

I've found dropping down to C to be a pain to set up the first time in any language (Ruby, Tcl, etc...) but once you have the tooling, it's just another workflow. When you think about it, even JavaScript takes a lot of setting up these days (preprocessors, linters, endless debates about why coffeescript is still a good language (hint: it's the existential operator)). But, you amortize the cost of this setup over the lifetime of the project so they are actually quite small.

But, it does seem that there are not regular blog posts about how to drop into C to make even the worst scripting language performant. Seems like maybe it's a dying art???

Dropping down to C is pretty cool in a language like Tcl. To toot my own horn, a bit, I worked on those chapters here: http://amzn.to/1U6sFPN

Erlang is a bit different though: if you're really concerned about being 'robust' with it, you don't want to link in some random C code in the main node. Not only could it crash, it might simply block for an unacceptable length of time, messing up the system's internal scheduler and making it unresponsive to events.

So with Erlang, to do things properly, you need to get the architecture right as well, which makes it that much tricker.

I think compared to most FFI API's the NIF interface is extremely clean and straight forward.

See my other reply though: it's not just about the API, which is nice, but about getting the architecture right. It does make the system more robust, but it requires more thought and planning.

Sure. Once you cross that barrier, you now have unlimited ammo and all the feet that ever were or will be, so you have to be a lot more diligent about what you're doing.

I guess my point is that in, say, Tcl, you pretty much just worry about calling out to the C code and if it takes 2 seconds to process an image or whatever, so be it. You merrily go about your business once the call returns. In Erlang there's more to think about.

I think that's technically true in Erlang as well... it's just that it makes all the other stuff that people expect from Erlang to mostly stop working smoothly. :-)

Reply on news.ycombinator.com