[thelazydogsback]

Some of this reminds of me a combination Icon and Mozart. (And happened to just come across my old Icon book a few days ago.)

Maybe the backtracking concepts are "too built in?" - e.g., (1|2) is not a first-class/reified "amb" object, right? - so if I want to introduce a different search than depth-first backtracking (breadth, dependency-directed), etc., I couldn't directly.

Seems like there could be some confusion between multiple, backtracked values and actual sequences -- this often leads to sub-optimal/confusing Prolog code as well, deciding when you have an explicit control structure vs. backtrack, when to "harden" results using setOf/bagOf/findAll, etc.

[mncharity]

Can anyone speak to this for Icon?

For context, Prolog had a culture of "the built-in search is less than wonderful... but good for crafting a problem-appropriate search". II(fuzzily)RC, Mozart/Oz decoupled nondeterministic code (explicit amb, spelled "dis") specification from search, and allowed managing search over nested spaces of unification constraints and threads. With (very fuzzy) less use of backtrack-driven implicit sequences than in Prolog? I so don't remember Icon.

Curiously, https://www2.cs.arizona.edu/icon/books.htm includes a "newish" (2010) book, "Icon Programming for Humanists".

[cryptonector]

In Icon a function ("procedure") can: fail, return a value, or "suspend" (yield) a value (and then again and again). The difference between returning and suspending being that a generator that returns cannot be resumed again. That difference was needed (IIRC) because the alternative would be to end the generator with failure, which could be confused with failure in a boolean sense. jq gets this better by saying that a function can produce zero, one, or more values (just like Verse), but unlike Verse, jq does have booleans.

I prefer the jq approach to the Icon approach, which I think means I am suspicious of this aspect of Verse :)

Also, boolean values are inherently useful, and while checking that an expression is empty or not is a boolean predicate, it should be a predicate that produces a boolean value. Otherwise if we have no boolean values then we shouldn't have integer values either and just go back to the Lambda Calculus!

[chowells]

Boolean values are inherently a sign of an insufficient data model. Booleans carry no inherent meaning. They don't tell you where they came from or provide any context about what operations or data they are guarding.

If your language lets you declare your own algebraic data types, there's no need to lean on a built-in Boolean type. You can create types that actually carry all the information you need so that pattern matching provides provenance and data only in the branches that need them.

Robert Harper discusses the idea in some detail here: https://existentialtype.wordpress.com/2011/03/15/boolean-bli...

[charcircuit]

Pattern matching is just syntatic sugar over using booleans and extracting values from data types. This is an exaggerated problem with a cherry picked solution. See what happens when the condition is if a number > 100. Are you expected to convert the integer into a church encoded integer and then write 100 S types? Are you expected to adopt a language with refined or dependent types? Adding a ton of abstractions to fix a problem doesn't mean you now don't have an even bigger problem.

[chowells]

Why do you care if a number is greater than 100? That has to actually mean something to you. What does it actually mean? Is it a validation error? Is it a warning trigger? Why not have a data type that actually expresses that meaning?

Wait, are you hung up on implementation nonsense? Sure, there's a low-level operation that returns a 0 or 1 that you convert into your actually-meaningful data type. But that doesn't need to have a surface language boolean type that's privileged by the language design. And that's what's important here - language design. People write programs in a language, not compiled code. The model the language provides is of critical importance. It's how the users of the language think and how they communicate with each other.

And this isn't abstraction. Boolean is the abstraction. It throws away all but a single bit of information. This is creating data types with semantics that carry far more than that single bit. It's making your code less abstract.

[charcircuit]

>Why do you care if a number is greater than 100?

You could make a battle royal game well where people start taking damage when they are 100 units away from the center of the map.

>Why not have a data type that actually expresses that meaning?

It would add unneccessary complexity.

>Wait, are you hung up on implementation nonsense?

The base of what it being abstracted over is not nonsense.

>But that doesn't need to have a surface language boolean type that's privileged by the language design.

If exposing booleans lets people write simpler code then it makes sense to expose it instead of forcing complexity onto everything.

>Boolean is the abstraction.

No, booleans are a concept that the processor understands. Zero is false and nonzero is true. The processor has instructions like like logical and that can operate on these booleans. Branches can happen based off a bit in the status register. A bit has two states which maps to true and false. Booleans are a fundamental concept that is being built upon. The processor has a limited number of data types that it understands. Creating new data types is creating abstractions because those data types do not actually exist. You are simply hiding an actual data type in the new one you are creating.

[mrkeen]

> Boolean values are inherently a sign of an insufficient data model. Booleans carry no inherent meaning.

Would you say the same about Ints or Strings?

[HelloNurse]

Practical language-agnostic example: a database for a business where some VARCHAR(34) columns are IBAN codes and others are messages to be displayed on a 2 by 17 characters LCD screen.

They are two completely different data types that happen to have identical approximate representations but should be distinguished very strictly in a sufficiently expressive model.

For instance, it should be possible to turn an IBAN into a device screen message but not the opposite, but you can't do it if they are all "strings"; and the two types have different value constraints while a generic VARCHAR has none.

[mrkeen]

> They are two completely different data types that happen to have identical approximate representations but should be distinguished very strictly in a sufficiently expressive model.

About bools: that's my problem with this. If-statements now operate on anything, instead of just bools. And I don't think that makes sense, unless cheese.

About strings: It's great you can statically differentiate between their types. But if strings don't exist, it makes it a little hard to write WHERE clauses.

[mjburgess]

They're saying you should be able to define your own

The domain of int/str is infinite so there's less sense in redefining a number system for every property you want to model

For small finite domains it makes sense to define your own in terms of the problem domain

In this sense all two-valued types are bools

[Tarean]

I was wondering if they are planning to use some datalog-style compilation strategy, where all choices are dumped into b-trees and nesting is replaced by indexed tables which are joined.

But I think mixing this with unification vars is sort of an open research problem?

Plus Haskell had a `data-parallel` mode which used a similar compilation strategy. It was dropped for being very complex and having virtually no users. Plus using too much broadcasting and compiling higher order functions into vectorized code had a serious performance tax.