Hacker News Reader

Comment by vjeux | Hacker News Reader

We posted about some of those in the blog over the years. http://skiplang.com/blog

Some of the interesting ones:

- An overview of how memoization works and the MVCC model behind the scenes: http://skiplang.com/blog/2017/01/04/how-memoization-works.ht...

- How pattern matching is implemented and the tricks to make goto work in JavaScript: http://skiplang.com/blog/2017/11/15/simulating-goto-in-javas...

- The work done on making error messages much more helpful by understanding common idioms from other programming languages: http://skiplang.com/blog/2017/11/20/fixing-the-syntax-barrie...

- The macro syntax that elegantly solves a lot of use cases where dynamism is commonly used http://skiplang.com/blog/2018/07/24/macros.html

View on HN

Thank you, I feel like Skip is very close to the language I've always wanted. Its MVCC looks very similar to Clojure's, I'm having trouble finding a good link about it:

https://sw1nn.com/blog/2012/04/11/clojure-stm-what-why-how/

The gist of it is that instead of tracking multiple separate locations in memory for values, the software transactional memory (STM) references data by value. So if you assign the value {a: 42, b: 24} to two variables x and y, that value is only stored in memory in one place that the variables both point to. Then if one of the variables changes something, for example y.b = 25, this big tree structure works like copy-on-write and makes copies of branches when mutations occur. So internally, a: 42 is one reference and b: (24 or 25) is another reference. So rather than using 4 cells of memory, we've only used 3.

This frees the developer from having to micromanage memory resources and makes a lot of other things like concurrency "just work" without locks. Do I have this correct?

Your explanation looks correct!

The practical application is having a webserver that runs a bunch of requests in parallel but that can all share the same memoization cache. We don't want to lock the full memoization cache and therefore block all the other requests when one thread writes a new value.

FWIW, MATLAB does a lot of this kind of optimization in its internal implementation of the core datatypes. There are a lot of smart compiler/JIT people that have worked hard on it over the years to optimize the internal guts despite the naive design of the language due to its age and some poor design decisions made ages ago that cannot be changed due to extreme backwards compatibility requirements. Kind of like javascript and V8/SpiderMonkey :)

Thank you, I didn't know that MATLAB did that. I'm a huge fan of it and hope that someday other matrix programming frameworks like TensorFlow, OpenCL and CUDA move to a similar general-purpose syntax that frees the developer from having to deal with the boilerplate associated with old OpenGL-style buffer representations.

Reply on news.ycombinator.com