Compact Linear Types¶
A compact linear type either the unit type 1, the void type 0, or any small sum or product of compact linear types. Small here means the representation as described below fits into a 64 bit machine word.
Compact linear types have two primary uses: for sub word level value manipulation and to support polyadic array operations.
The following types are unitsum types:
typedef void = 0; typedef unit = 1; typedef bool = 2; // 1 + 1 4 // 1 + 1 + 1 + 1 5 // 1 + 1 + 1 + 1 + 1 ...
These are sums of 0, 1, 2, 3, 4, 5 .. units and represent 0, 1, 2, 3, 4, 5 alternatives. Values of unitsum types have two notations:
case 3 of 5 `3:5
The first syntax is general, the second is peculiar to unitsums. Note that there are no values of type void, because there are no alternatives. The value of type unit can also be written:
and represents exactly one alternative. Values of type bool can also be written
false // `0:2 true // `1:2
Note that somewhat unfortunately, the value index is zero origin so case 0 of 5 is the first of 5 cases and case 4 of 5 is the last. It reads badly but zero origin was chosen for symmetry with C array indexing conventions.
All unitsums from 1 up are represented by 64 bit unsigned integers. [This may be relaxed and/or extended in future versions of Felix]
Products of compact linear types are compact linear types. For example:
var x : 2 ^ 4 = true, false, false, true; println$ x.1; // false
This is an array of 4 bits, but it has a magic property: it is compact linear so it is represented by a single machine word. Arrays of up to 64 bits are represented by single machine words. Here is another example:
var x : 2 * (3 * 4) = true, (`1:3, `2:4); println$ x.1.1._strr; // case 1 of 3
Again, the value is a single machine word. Compact linear types are similar to C bit fields, however a C bitfield must consist of $n$ bits and so represents $2^n$ values. Felix compact linear types have no such constraint.
Compact linear types are represented by standard variable radix number system. The easiest explanation is the following example:
var time = (`2:24, `30:60, `1:60); // 1 second past 2:30 am var secs = time :>> int; println$ "Seconds into day = " + secs.str; assert secs == 60 * 60 * 2 + 60 * 30 + 1;
Compact linear types are named that because if there are $n$ possible value they’re represented by a range of integers from 0 upto $n-1$. This range is compact, meaning there are no holes in it, and linear, because it is integral.