Hacker News Reader

Comment by djacobs | Hacker News Reader

I don't get the appeal of coding like this. To me, a unary operator shouldn't mutate its receiver. (Edit: neither should a binary one.) Instead, the resulting value is meant to be assigned to a new value or otherwise manipulated.

This is a sort of DSL, but it doesn't really give us any more clarity than traditional allow/disallow arrays, does it?

It's this:

  + url("your-url-here")

versus this:

  allowed << url("your-url-here")

Why is the former more attractive?

View on HN

As a programming construct, it isn't.

As a configuration language, it might be.

Bingo. It's executable code that reads like a config file, that's about as good as configuration can get.

What's the use case, though? If you want config file syntax, use a config file. If you want a program, use code.

Building up an object system like this for simple configuration seems excessive to me.

I would do something like this:

  c = YAML.load_file 'config.yml'
  allowed, disallowed = c['allowed'], c['disallowed']

YAML is kind of confining as a config file language.

> What's the use case, though? If you want config file syntax, use a config file. If you want a program, use code

Data is code is data, avoids having to write a parser (or even reuse an existing one), avoids having to resolve the impedance mismatch between the programming language and the configuration language, scales up infinitely in case the user needs more power (which is not that rare, especially for a rules system which is the use case here)

> I would do something like this:

> c = YAML.load_file 'config.yml' > allowed, disallowed = c['allowed'], c['disallowed']

Now you have two superfluous lines in your code and you have not gained anything of value.

I don't have any superfluous lines in my code. I wanted to load a config file into my runtime, and I did. Not superfluous at all. The (extraordinarily) superfluous code is the code used to set up the object config system where none is needed.

> Data is code is data.

This is hardly a case of homoiconicity, is it?

> Not superfluous at all. The (extraordinarily) superfluous code is the code used to set up the object config system where none is needed.

You need that code to load/create rules in any case, but for you it's done from a config file which the user writes. For TFAA, the user loads his data directly into the running system. Therefore the layer you have to setup between the config file and the actual running code (parsing the config file then interpreting its values to translate them into a running ruleset) is superfluous: you can do without it.

Because you know, your "example" YAML parsing is useless in and of itself, it doesn't result in a working ruleset, you skipped the whole part where you feed that data to your code and setup the stuff that actually does work.

TFAA removes the parsing and interpretation from the chain and instead lets the user feed data directly to the running code without layers inbetween.

  Because you know, your "example" YAML parsing is 
  useless in and of itself, it doesn't result in a 
  working ruleset, you skipped the whole part where you 
  feed that data to your code and setup the stuff 
  that actually does work.

Sorry, but that's not actually the case, is it? There is no "business logic" in his example. To actually use the config system he's set up, you need to evaluate Rule#charge for every rule you want to evaluate. In my example, all you could run allowed.include? or simply iterate over the allowed array.

Homoiconicity has nothing to do with the Church-Turing Thesis.

____

Church: you can represent data using functions/function application.

    Historic example: Church encoding of the natural numbers. 
    Day to day example: The hash table and the array are in fact pure functions.

____

Turing: the input of a machine (data) can be used to carry computations.

    Historic example: the Turing machine (duh :-)
    Day to day example: from machine language to Coffeescript.

It's often handy to have a full blown programming language at your disposal in a configuration file. Want to pull a configuration setting out of another file? Want to loop through things to avoid repetition? Want to switch some but not all configuration settings depending on some parameter? Depending on which hardware you're running on? Easy.

> To me, a unary operator shouldn't mutate its receiver. Instead, the resulting value is meant to be assigned to a new value or otherwise manipulated.

Why? And why would you draw that distinction for unary operators but not binary ones? Could it be just out of habit, because you've seen binary operators overloaded to mutate operands before but have never seen that with unary ones?

If the language implements unary operators as regular methods/messages (rather than special forms) why couldn't they be overloaded in the same way binary operators are, as long as it makes sense and reads correctly? (and in this case it certainly does)

And why would you draw that distinction for unary operators but not binary ones?

I wouldn't -- I think it's wrong for binary operators too.

Fair enough, but that most definitely isn't djacobs's case considering the replacement snippet he wrote.

I don't really want to speak for him, but what I got from his post was that he was saying that the unary syntax is as bad as C++'s (imho retarded) cout syntax.

Better than either of those, I think, would be

  allow url("your-url-here")

(with reject or something similar for the opposite).

You could even drop the url method and simply write

  allow "your-url-here"

if you only need to allow/disallow based on URLs.

Funny that you argue against one overloaded, mutating operator with another overloaded, mutating operator that has been stretched even further from its original meaning.

Reply on news.ycombinator.com