Mirror of forum.nim-lang.org

6850 :: stdlib pattern matching feedback

• haxscramper_ (orginal) [2020-09-23T13:39:33+02:00] view original

  I and @alehander42 started working on adding pattern matching in stdlib. Right now we at ideas brainstorming stage - what is needed, syntax, potential escape hatches for user-defined behavior etc. There have been some discussion in RFC on github - link and on discord/IRC.

  Current syntax/planned features are described in linked comment, but mostly feedback is needed for following questions:

  • Would you rather prefer DSL to introduce keywords (add, until, with) or work with special symbols like ?@, *@, .._? I'm more in favour of using keywords instead of symbols.
  • Your use cases. It is clear that matching NimNode is a very common task, but there might be something else that occurs quite often, like json data extraction for example.
  • Any ideas/suggestions about behaviour/syntax of pattern matches

  Some examples of current syntax:

  There is a proof-of-concept I'm working on right now: examples of use for it can be found here . It uses *@ syntax, but I will most likely switch it to keyword-based one.

---

• spip (orginal) [2020-09-23T13:58:20+02:00] view original

  Have you had a look at the syntax proposed by the different pattern matching modules?

---

• haxscramper_ (orginal) [2020-09-23T14:06:17+02:00] view original

  Yes. I adapted case object idea from patty and some syntax elements from gara. ast-pattern-matching only supports nim node matching so it is not really applicable in general case of pattern matching.

---

• elcritch (orginal) [2020-09-24T17:31:39+02:00] view original

  Pattern matching is one of my favorite aspects of Elixir. It'd be fantastic having it in the Nim stdlib :-) Out of curiosity, how would this compare to that new Python PEP?

---

• sschwarzer (orginal) [2020-09-25T00:03:31+02:00] view original

  For the others, the PEP is this.

---

• haxscramper_ (orginal) [2020-09-25T13:31:10+02:00] view original

  It is really similar to the python version (e.g. a lot of similar features and syntax). Sequence pattern matching is more feature-full, with support for things like all/none/any - closely matching templates from sequtils module (namely allIt and anyIt), so you should be able to write things like

  case someInputJson:
    of [all @elem is JString()]:
      echo "all ok"
    of [any @elem is JString()]:
     echo "that's worse, but we can deal with it")
    else:
      raiseAssert("Expected at least some elements to be json strings !")
  Run

---

• Araq (orginal) [2020-09-25T14:07:06+02:00] view original

  Beware of any, people will use it and then need to know which element really did match. And if that's not available they have to use find afterwards. And then we have the typical performance vs convenience tradeoff. But it's a tradeoff Nim doesn't have to make with its DSL powers.

---

• haxscramper_ (orginal) [2020-09-25T14:20:00+02:00] view original

  Currently any captures all elements in @val, not just first one. I agree this might be confusing, so maybe renaming it to findAll and find might be a good idea.

  Indices of matched elements are not accessible in any predicates right now, but that could be fixed by injecting matches: seq[int] in the scope. Do you think it is necessary to do this?

---

• Araq (orginal) [2020-09-25T17:23:24+02:00] view original

  but that could be fixed by injecting matches: seq[int] in the scope. Do you think it is necessary to do this?

  Why a seq? any can stop after the first found match.

---

• haxscramper_ (orginal) [2020-09-25T17:57:38+02:00] view original

  It might be a question of terminology mostly, but in my mind

  • all means "every element without exception" (replace with each, every ?)
  • any - at least one, maybe more, test all elements in sequences regardless
  • none - none. Inverse with all.
  • first or find - at least one, stop after first found element.

  On input sequence [1,2,3]

  • [any @elem.(it < 3)] should give elems: seq[int] = @[1,2]
  • [first @elem.(it < 3)] creates elems: int = 1

  So any/all will generate sequence of all elements that match predicate, mostly working like filterIt. Using "filter" as a keyword is not applicable here, because it does not reflect the fact we are testing for patterns and not transforming input.

---