Expected behaviour of push/npush/npush_if

[stdweird]

While working on #102, i stumbled on some ill defined behaviour of push (which also exists in npush and push_if)

object template push;

include 'pan/functions';

"/functions/push_dml_empty/data" = {
    foreach(idx;id;list(1,2,3)) {
        push(id);
    };
};

"/functions/push_dml_nonempty/data/0" = 0;
"/functions/push_dml_nonempty/data" = {
    foreach(idx;id;list(1,2,3)) {
        push(id);
    };
};

current master returns

{
  "functions": {
    "push_empty": {
      "data": [
        3
      ]
    },
    "push_nonempty": {
      "data": [
        0,
        1,
        2,
        3
      ]
    }
  }
}

so there is a significance difference between push depending on the the existence of SELF (the reimplementation using merge as in the PR suffers from similar yet different problems. (chagning the code and ending with SELF; gives different issues altogether, but they are still different).

my question is: how to handle this. keep the old code, but warn in the doc that push/npush/push_if have undesired sideeffects in DML blocks?

some sanity could be restored by using append in push/push_if, but there is no equivalent of append for dicts (in particular, merge with single argument to update SELF is not supported; see quattor/pan#103).

[jrha]

We agreed at the 21st workshop that the observed behaviour is not desired or intentional.

[stdweird]

we will fix this issue as a bug-fix in the core templates (probably by using append).
but we will not add push as a buitlin (see quattor/pan#108 for that discussion)