Mirror of forum.nim-lang.org

8978 :: any particular reason on deprecating IndexError?

• haoliang (orginal) [2022-03-05T00:54:27+01:00] view original

  when i get value from array/deque, i prefer the EAFP way. it is also possible in nim, but the IndexError has been deprecated, and IndexDefect is designed to be uncatchable. i found this issue, but could not found any particular reason; i know "look before you leap" way to get thing done, and i do it in my little gist.

  so i really wonder why IndexError had been deprecated, especially in my opinion IndexError is similar to KeyError.

  i asked this question in IM channel, but i did not get good answer before the question has been flushed away.

---

• ElegantBeef (orginal) [2022-03-05T01:27:24+01:00] view original

  My view is that exceptions are for exceptional code, not for simple logical errors. As such anything that can be prevented by simple checks in your code should not be caught but rather handled with checks(in the case the checks are cheaper and not complex). KeyError is different than IndexError for the complexity reason mentioned, to check if an index is valid it's a simple a in typeof(arr).low .. typeof(arr).high whereas a checking if a key is in a table is a hash of the type, which can be relatively expensive. Then there are also procedures that let you avoid a KeyError. Defects inform the programmer they have a logic error that can be prevented easily and cheaply, exceptions inform the programmer there is something out of their control causing an error.

---

• sls1005 (orginal) [2022-03-05T05:02:45+01:00] view original

  Sometimes Nim programmers use -d:danger or --checks:off to speed up their code. The boundary checks would no longer be performed, and the only thing that can prevent overreading would be manual checks. That's why a index out of range is called a "Defect."

---

• haoliang (orginal) [2022-03-05T11:56:32+01:00] view original

  unfortunately, ElegantBeef's reply did not really convince me.

  it's good to know the use of -d:danger and --checks:off with this scenario.

  i tried compile a snippet with --checks:off, the output surprised me: deque.popLast() always returns 0; does that make sense? there is no document describes this behavior.

  
  proc test_deque =
    var x = [1].toDeque
    
    echo "1: ", x.popLast
    try:
      echo "2: ", x.popLast
      echo "3: ", x.popLast
      echo "4: ", x.popLast
    except IndexError:
      echo "catched IndexError"
  
  
  when isMainModule:
    test_deque()
  Run

  in the end, i think that behavior is a good reason to keep away from EAFP in nim.

---

• haoliang (orginal) [2022-03-05T12:04:11+01:00] view original

  My view is that exceptions are for exceptional code, not for simple logical errors.

  even Araq will break this rule, see

---

• arnetheduck (orginal) [2022-03-05T12:23:17+01:00] view original

  IndexError was always a Defect - it merely changed name so as to highlight that it's not catchable (and never has been).

  The choice between Defect and Error is, like you're noticing, quite arbitrary - there's no line that one can draw between them, rather it depends on the mood of the developer at the time of writing and sometimes the surrounding code. It is also quite subjective what you consider "exceptional" and not, so that's not that much of a useful distinction, except (haha) in extreme cases.

---

• planetis (orginal) [2022-03-05T12:35:54+01:00] view original

  when i get value from array/deque, i prefer the EAFP way.

  BTW simple benchmarks in the past show that raising exceptions is slower than checking for key.

---

• Araq (orginal) [2022-03-07T07:47:20+01:00] view original

  > My view is that exceptions are for exceptional code, not for simple logical errors.

  even Araq will break this rule, see

  Correct, exceptions are control flow so I use them for control flow.

---

• Araq (orginal) [2022-03-07T08:58:25+01:00] view original

  The choice between Defect and Error is, like you're noticing, quite arbitrary - there's no line that one can draw between them, rather it depends on the mood of the developer at the time of writing and sometimes the surrounding code.

  Actually, the design principle is "IndexDefect is for arrays/seqs and for seq-like containers" (which a Deque is).

  And yes, this is "inconsistent" with Hash tables that raise KeyError instead. Hash tables are not used for algorithms like a[x..y] = b[x..y] * c[x..y]. Different use cases produce different programming patterns which produce different takes on the error handling philosophy.

---