Index Generator


 * This page is about generating indices for an array of a given size. See Indexing, Indices, Index of, and Interval Index for other operations named after indices.

or Interval, often called by the name of its glyph Iota, is a monadic primitive function which returns an array of indices with shape given by the right argument. In the result, the element at each index is that index.

Originally, Iota was defined only on a single number. In APL/360, and all later APLs,  for a scalar   returns a simple numeric vector of length   counting from the index origin up. In nested APLs, this result may be seen as an array of scalar indices; NARS extended Iota to allow a vector argument by making the result for a non-singleton vector be an array of vector indices. Flat APLs do not use this extension, but may extend to multiple argument elements in different ways. A Dictionary of APL defines Iota to have function rank 0, and SHARP APL gives it rank 1 but requires each row (1-cell) of the argument to have length 1. In A+ and J Iota of a vector returns an array which counts up in ravel order, that is, it matches Iota of its bound, reshaped. A+ refers to this function as Interval while J calls it Integers.

Examples
returns the first  indices in order: This result depends on index origin: if the index origin is set to zero, it starts counting from zero instead: An argument of  produces the empty vector, Zilde:

Vector arguments
In A+ and J, Iota always returns a simple array. The equivalence     defines the result for non-singleton vectors. If the argument is an empty vector, the result is a scalar: the index origin (which is always 0 in these languages). We might call such a function a "ravel index generator" since the result is an array of scalar indices into its own ravel. In contrast, nested APLs, if they implement Iota for a vector argument (APL2 and APLX, for example, do not), use indices of elements: The result of  should have shape , implying it is a scalar, and its only element must be the only possible index into a scalar, the empty vector. This means it must be. However, Dyalog APL returned the scalar  instead prior to version 13.0.

Negative arguments
Two languages extend Iota to allow negative arguments. In both cases, the shape of the result is the absolute value of the argument.

In the nested APL NARS2000, negating one number in the argument subtracts that value from each index along the corresponding axis.

In J, a negative number reverses the result along that axis.

Scalar-vector discrepancy in nested APLs
In NARS and later nested APLs which share its definition, the result of Iota is simple when the argument is a singleton and nested otherwise. This is because, when the argument is a singleton, each index in the result is represented as a scalar rather than a 1-element vector. dzaima/APL and Kap break from this definition by using vector indices whenever the argument is a vector, and scalar indices if the argument is a scalar, so that the result is only simple if the argument was scalar. This definition has the property that the shape of each index matches the shape of the argument.

Such an extension breaks compatibility with earlier non-nested APLs: although these APLs required the argument of Iota to be a singleton, they always allowed a scalar or vector, and sometimes any singleton. This is because a shape to be passed to Iota would often naturally be a vector: for example, Shape always returns a vector, and a function such as Tally to return a scalar length did not exist.

NARS defines its extension using the equivalence     based on an Outer Product reduction. For a singleton vector, this does yield a simple result, because   leaves a singleton argument unchanged. However, this is arguably due to inconsistent design in Reduce: NARS's creator, Bob Smith, argues that Catenate reduction should yield a DOMAIN ERROR, and the same arguments apply to its Outer Product.

History
The Greek letter iota, corresponding to the Latin letter i, now usually treated as a mnemonic for "integers" or "index", more likely stood for "identity" in early Iverson notation. Iverson's 1960 paper The Description of Finite Sequential Processes states that "the identity permutation vector $$\iota$$ is defined by the relation $$\iota_j=j$$." Since $$e$$ is a more typical letter to use for the identity permutation, it's possible that Iverson's choice was also motivated by the convention of using $$\iota$$ for an inclusion map, where iota now indicates "inclusion" or "injection". The term "inclusion" typically indicates that a particular injection is meant to be treated as an identity.

The notation $$\iota_o(n)$$, "Interval", was used in A Programming Language for the vector of the first $$n$$ indices starting at index origin $$o$$. Both arguments were optional, with $$n$$ being implied by conformability if omitted. It was adopted with the obvious alterations in APL\360, and extended to a vector argument for nested APLs by NARS and for flat arrays by A+.

Outside of APL
The name "iota" has sometimes been used to indicate an increasing sequence of integers even in languages other than APL. In the C++11 standard library,  fills an iterator with such a sequence, and was named after the APL glyph. The Go language also uses the name  as a predeclared identifier which represents increasing integers: each time it is used within a single constant declaration its value will be one higher. Functions named "iota" appear in libraries such as Scheme's SRFI 1 (list library), GPGPU library ArrayFire, and JavaScript library scijs. The BSD operating system (including derivative macOS) uses  to print sequential data, a name derived in part from APL's glyph.

Tutorials

 * APL Cultivation

Documentation

 * Dyalog
 * NARS2000
 * APLX
 * J Dictionary, Nuvoc
 * BQN (as Range)