Errors in Reference Manual

[stratosn]

reporter issue reference	document (UM/RM/EBNF)	page	line
DI-21	RM	162	if-then-else p. 162
DI-22	RM	162	if-then-else p. 162

---

Keyword elseif is unnecessary. Construct else if does the trick without any syntactical complications with elseif, because after else one can put another if. Sub-expressions ds_1, ds_2 and ds_3 are incorrectly described as either datasets or constants. Any expression returning either a dataset or some other VTL 1.1 type (scalar, etc.) can be used. For instance, (a+4*b+c)/6 is a scalar expression which is perfectly valid to stand after then or else but is not a constant.

Expain what is the type of if A then B else C based on types of B and C.

[capacma]

I agree to eliminate elseif. NB: VTL defines if-then-else as an expression but in most programming languages if-then-else is a statement and not an expression. It would be wise to replace the if-then-else syntax with the "case" syntax that can be found in many programming languages: case when Condition then Expression { when Condition then Expression} * else Expression end

[bellomarini]

• I agree to eliminate elseif, which is useless.
• I also agree that in general a multi-option form would be better.
• No particular problems with case when: we could even have both as it is mostly personal style in programming.

Independently of the syntax we choose, this conditional operator should remain an expression.

Since we are in a declarative language, we cannot have it exactly as a control flow control flow operator. However, I understand that a control-flow-like version is easy to understand and intuitive to many users, thus I would suggest the introduction of a pattern matching operator, syntactically more compact (yet semantically equivalent to the expression form of case-when) and used in functional languages, but "visually" resembling an imperative flow control structure.

An example follows:

DS1 =  match DS2 with:
| DS3 -> DS2*2 + DS3
| DS5 -> (DS2 + DS5) / 2
...

• with the constraint that DS2, DS3, DS5 all have the same identifiers and measures

For all the datapoints of DS2 that have correspondences in DS3 (in terms of identifier components), return a Datapoint calculated by multiplying the measures by 2 and summed to DS3. For all the datapoints of DS2 that have correspondences in DS5, return (DS2 + DS5)/2 and so on.

Nested expressions in the pattern matching operator can give a "visual" idea of flow control, though remaining in the functional and declarative paradigm.

DS1  = match DS2 with:
   | DS3 -> match DS4 with:
       | DS5 -> DS4 + DS3
       | DS6 -> DS4 - DS3
   | DS2 -> match DS2 with:
      | DS7 ...