6 ms·

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 a

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 4y ago
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 https://www2.cs.arizona.edu/icon/books.htm includes a "newish" (2010) book, "Icon Programming for Humanists".
- cryptonector 4y ago
  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 4y ago
    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-blindness/ https://existentialtype.wordpress.com/2011/03/15/boolean-bli...
    - mrkeen 4y ago
      > 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?
      - mjburgess 4y ago
        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
      - HelloNurse 4y ago
        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.
        4y ago
Tarean 4y ago
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.

mncharity 4y ago

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

cryptonector 4y ago
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 4y ago
  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-blindness/ https://existentialtype.wordpress.com/2011/03/15/boolean-bli...
  - mrkeen 4y ago
    > 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?
    - mjburgess 4y ago
      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
    - HelloNurse 4y ago
      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.
      - 4y ago

Tarean 4y ago

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.