Mirror of forum.nim-lang.org

6809 :: Add hash[T] to hashes?

• alexeypetrushin (orginal) [2020-09-13T12:57:38+02:00] view original

  Currently tables module require the hash proc to be written manually if the object used as a key.

  Why not define default hash function for objects? Something like:

  proc hash*[T: tuple|object](o: T): Hash =
    var h: Hash = 0
    for f in o.fields: h = h !& f.hash
    !$h
  Run

---

• sschwarzer (orginal) [2020-09-13T13:14:53+02:00] view original

  I like the idea of having a default implementation for objects.

  The default implementation should also consider object variants. Related: https://forum.nim-lang.org/t/6781 .

---

• Araq (orginal) [2020-09-13T15:23:29+02:00] view original

  Object identity cannot be automated.

---

• sschwarzer (orginal) [2020-09-13T16:18:17+02:00] view original

  Object identity cannot be automated.

  I'm not sure I understand what you mean. As I understand the OP, this is about equality (on a business domain level), not identity. Also, I speak about a default. Of course, it should be possible to override the default implementation with a custom implementation. That said, if the default implementation would be inapplicable in most situations, having the default maybe causes more problems than it solves. Is that what you meant?

---

• Araq (orginal) [2020-09-13T18:13:35+02:00] view original

  Well guessing that object means tuple isn't good enough for me. I keep getting (minor) bugs with $ that "simply exists" for my objects which are not trivial, so yeah, these defaults cause problems.

---

• slonik_az (orginal) [2020-09-13T20:03:42+02:00] view original

  In conjunction with tables module having default hash function implementation is a bad idea. For hash-table consistency keys should be immutable, so that their hashes do not change over the lifespan of the key objects. With a fallback hash implementation applicable to any object this constrain is impossible to enforce.

  --Leo

---

• sschwarzer (orginal) [2020-09-14T00:34:02+02:00] view original

  In conjunction with tables module having default hash function implementation is a bad idea. For hash-table consistency keys should be immutable, so that their hashes do not change over the lifespan of the key objects.

  We already have this situation with the default hash implementations in the hashes module:

  import tables
  
  var t = Table[seq[int], int]()
  
  var key = @[1, 2, 3]
  t[key] = 3
  echo t[key]
  key.add(4)
  echo t[key]
  Run

  https://play.nim-lang.org/#ix=2xsB

  resulting in

  
  3
  /usercode/in.nim(9)      in
  /playground/nim/lib/pure/collections/tables.nim(262) []
  Error: unhandled exception: key not found: @[1, 2, 3, 4] [KeyError]
  Run

  Client code is responsible for not changing keys.

---

• alexeypetrushin (orginal) [2020-09-14T05:28:18+02:00] view original

  Agreed.

  A possible improvement would be to add autohash helper to hashes module. So we avoid automating object identity and making it easier for people to auto generate hashes for cases when it make sense.

  
  proc autohash*[T: tuple|object](o: T): Hash =
    var h: Hash = 0
    for f in o.fields: h = h !& f.hash
    !$h
  
  proc hash*(v: Something): Hash = v.autohash
  Run

---