Mirror of forum.nim-lang.org

7135 :: tables.add() deprecation in favor of []= is wrong

• mildred (orginal) [2020-11-21T17:43:17+01:00] view original

  With Nim 1.4, there is a deprecation warning Warning: Deprecated since v1.4; it was more confusing than useful, use `[]=`; add is deprecated that I believe is wrong. []= and add are doing two very different things for tables.

  • add is adding a new pair to the table, which may result in duplicated values for the same key
  • []= is replacing the existing value for the same key if it exists

  When handling HTTP parameters, I need to have tables with duplicate values for the same key, and I'm using add for this but I get the warning. Is this deprecation really relevant, is there a replacement for add ?

---

• DeletedUser (orginal) [2020-11-21T17:59:51+01:00] view original

  Would seq[(string, string)] not suffice? I would assume you wouldn't need fast indexing and deletion since you dealt with multiple keys and were probably not already using them.

---

• mildred (orginal) [2020-11-21T18:25:25+01:00] view original

  Most keys are not duplicated, and fast indexing may help there. Only the odd key may need to iterate over the table to discover all its values.

  In fact, I got the impression that Tables were explicitly able to deal with duplicate keys, and I started using them for this purpose. If this is removed, perhaps a MultiTable type should be created.

  Well, I think I got my answer:

---

• mildred (orginal) [2020-11-21T18:32:39+01:00] view original

  Found some interesting discussion here: https://github.com/nim-lang/Nim/pull/15047

---

• sschwarzer (orginal) [2020-11-21T19:07:00+01:00] view original

  There was a discussion on this recently: https://forum.nim-lang.org/t/7055

---

• cblake (orginal) [2020-11-21T21:14:07+01:00] view original

  The in-stdlib replacement is Table[K, seq[V]]. The nimbleverse has at least adix/lptabz which tries to be close to the stdlib Table interfaces while extending functionality and configurability/flexibility in several dimensions.

---

• mildred (orginal) [2020-11-22T00:21:48+01:00] view original

  Note: the linked benchmark at https://github.com/nim-lang/Nim/pull/15047 are wrong. There is overhead to using seq[T] over a table: adding an item must perform a lookup to know if a seq exists (then append to it and store it) or not.

---

• cblake (orginal) [2020-11-22T00:33:20+01:00] view original

  You can always tab.mgetOrPut(key, @[]).add(val) to do a 1-lookup variant. There is, of course, memory indirection/allocation overhead.

---