Integer type is not defined

[Eswcvlad]

In a bunch of types integer is specified as the parameter type, while integer is not directly specified in the spec and there are no coercion rules for it.

I've done some testing and in this implementation, coercion behavior varies from function to function, when you pass a floating-point number:

• Truncate the fractional part:

  • datetime
  • eomonth
  • left
  • replace
  • rept
  • right
  • time
  • toString

• TypeError:

  • fromCodePoint

• Other:

  • deepScan -> looks like name just gets coerced to string instead.
  • find -> some strange behavior, find("m", "abm", 0.7) returns 2.7.
  • hasProperty -> looks like name just gets coerced to string instead.
  • mid -> thought it was truncation, but mid("Fluid Flow", 0.1, 4.9) returns Fluid instead of Flui.
  • round -> some unpredictable result, ex. round(1.95583, 3.8) returns 1.9557581994970148.
  • search -> similar to find, search("a?c", "acabc", 1.2) returns 1.2.
  • substitute -> not sure, what happens; substitute("aaaa", "a", "b", 1.2) returns "aaaa".
  • toNumber -> ignores value and uses 10 instead
  • trunc -> similar to round, ex. trunc(8.9122345, 4.1) returns 8.912203927959501.
  • value -> tries using number as-is, which results in null.
  • weekday -> returns null.

It would be better, if it was standardized across the board. Imo, converting to Number and then truncating the fractional part seems to be the way to go. TypeError doesn't look ideal for this, as it could cause issues in languages like JS, where numbers are treated as floats.

There are also no type size restrictions on integer either, but I assume it is not an issue, as, practically speaking, most reasonable values will fit [-(2^53 − 1), (2^53 − 1)] and there is little sense in using big integers for this, so it could be left up to implementations.

[JohnBrinkman]

Valid concerns. I will look at a fix.

[JohnBrinkman]

I don't think we can make integer a first-class data type in json-formula, since it's not a JSON data type -- and the usage doesn't occur outside of functions. I think it's best just to add some discussion in the Data Types section to describe the limited support.

[Eswcvlad]

I don't think we can make integer a first-class data type in json-formula, since it's not a JSON data type -- and the usage doesn't occur outside of functions. I think it's best just to add some discussion in the Data Types section to describe the limited support.

Makes sense. I was mostly concerned about inconsistent coercion.

[JohnBrinkman]

Update spec to say "truncate" instead of "floor"