The programming language APL is distinctive in being symbolic rather than lexical: its primitives are denoted by symbols, not words. These symbols were originally devised as a mathematical notation to describe algorithms.[1] APL programmers often assign informal names when discussing functions and operators (for example, "product" for ×/) but the core functions and operators provided by the language are denoted by non-textual symbols.
Most symbols denote functions or operators. A monadic function takes as its argument the result of evaluating everything to its right. (Moderated in the usual way by parentheses.) A dyadic function has another argument, the first item of data on its left. Many symbols denote both monadic and dyadic functions, interpreted according to use. For example, ⌊3.2 gives 3, the largest integer not above the argument, and 3⌊2 gives 2, the lower of the two arguments.
APL uses the term operator in Heaviside’s sense as a moderator of a function as opposed to some other programming language's use of the same term as something that operates on data, ref. relational operator and operators generally. Other programming languages also sometimes use this term interchangeably with function, however both terms are used in APL more precisely.[2] [3] [4] [5] [6] Early definitions of APL symbols were very specific about how symbols were categorized.[7] For example, the operator reduce is denoted by a forward slash and reduces an array along one axis by interposing its function operand. An example of reduce:
×/2 3 4
24
| << Equivalent results in APL >> << Reduce operator / used at left |
2×3×4
24
|
1 0 1\45 67
45 0 67
| << Opposite results in APL >> << Expand dyadic function \ used at left Replicate dyadic function / used at right >> |
1 0 1/45 0 67
45 67
|
In APL the precedence hierarchy for functions or operators is strictly positional: expressions are evaluated right-to-left. APL does not follow the usual operator precedence of other programming languages; for example, × does not bind its operands any more "tightly" than +. Instead of operator precedence, APL defines a notion of scope.
The scope of a function determines its arguments. Functions have long right scope: that is, they take as right arguments everything to their right. A dyadic function has short left scope: it takes as its left arguments the first piece of data to its left. For example, (leftmost column below is actual program code from an APL user session, indented = actual user input, not-indented = result returned by APL interpreter):
APL is going to execute from right-to-left.
Step 1) 4-5 = -1.
Step 2) 3 times -1 = -3.
Step 3) Take the floor or lower of 2 and -3 = -3.
Step 4) Divide 1 by -3 = -0.3333333333 = final result.
The diaeresis ¨ or mini double-dot means repeat or over each or perform each separately so iota repeats (in human i.e., reversed terms, the APL interpreter reads 3 3 over each use iota), concisely: iota for each 3.The left operand for the over-each operator ¨ is the index ⍳ function. The derived function ⍳¨ is used monadically and takes as its right operand the vector 3 3. The left scope of each is terminated by the reduce operator, denoted by the forward slash. Its left operand is the function expression to its left: the outer product of the equals function. The result of ∘.=/ is a monadic function. With a function's usual long right scope, it takes as its right argument the result of ⍳¨3 3. Thus (⍳3)(⍳3) 1 2 3 1 2 3 (⍳3)∘.=⍳3 1 0 0 0 1 0 0 0 1 1 2 3 1 2 3 ∘.=/⍳¨3 3 1 0 0 0 1 0 0 0 1
Equivalent results in APL: '''(⍳3)(⍳3)''' and '''{{font color|red|⍳¨3 3}}''' << Rightmost expression is .
The matrix of 1s and 0s similarly produced by '''∘.=/⍳¨3 3''' and '''(⍳3)∘.=⍳3''' is called an identity matrix.
Identity matrices are useful in solving matrix determinants, groups of linear equations and multiple regression.
Letters ⍳ FindIt3 1 2 6
First, variable Letters is assigned a vector of 5-elements, in this case - letters of the alphabet.
The shape ⍴ or character vector-length of Letters is 5.
Variable FindIt is assigned what to search for in Letters and its length is 4 characters.
1 2 3 4 5 << vector positions or index #'s in Letters
ABCDE
At left, dyadic function iota searches through its left argument(Letters) for the search string (iota's right argument, FindIt).
Iota finds letter "C" at position 3 in Letters, it finds "A" at position 1, and "B" at position 2. Iota does not find letter "S"
anywhere in variable Letters so it returns the number 6 which is 1 greater than the length of Letters. Iota found letters
"CAB" (3 1 2). Iota correctly did not find "S" (6).
| Name(s) | Notation | Meaning | Unicode code point | |
|---|---|---|---|---|
| Roll | One integer selected randomly from the first B integers | |||
| Ceiling | Least integer greater than or equal to B | |||
| Floor | Greatest integer less than or equal to B | |||
| Shape, Rho | Number of components in each dimension of B | |||
| Not, Tilde | Logical: 1 is 0, 0 is 1 | |||
| Absolute value | Magnitude of B | |||
| Index generator, Iota | Vector of the first B integers | |||
| Exponential | e to the B power | |||
| Negation | Changes sign of B | |||
| Conjugate | The complex conjugate of B (real numbers are returned unchanged) | |||
| Signum | ¯1 if B<0; 0 if B=0; 1 if B>0 | |||
| Reciprocal | 1 divided by B | |||
| Ravel, Catenate, Laminate | Reshapes B into a vector | |||
| Matrix inverse, Monadic Quad Divide | Inverse of matrix B | |||
| Pi times | Multiply by π | |||
| Logarithm | Natural logarithm of B | |||
| Reversal | Reverse elements of B along last axis | |||
| Reversal | Reverse elements of B along first axis | |||
| Grade up | Indices of B which will arrange B in ascending order | |||
| Grade down | Indices of B which will arrange B in descending order | |||
| Execute | Execute an APL expression | |||
| Monadic format | A character representation of B | |||
| Monadic transpose | Reverse the axes of B | |||
| Factorial | Product of integers 1 to B | |||
| Depth | Nesting depth: 1 for un-nested array | |||
| Table | Makes B into a table, a 2-dimensional array. |
| Name(s) | Notation | Meaning | Unicode code point | |
|---|---|---|---|---|
| Add | Sum of A and B | |||
| Subtract | A minus B | |||
| Multiply | A multiplied by B | |||
| Divide | A divided by B | |||
| Exponentiation | A raised to the B power | |||
| Circle | Trigonometric functions of B selected by A A=1: sin(B) A=5: sinh(B) A=2: cos(B) A=6: cosh(B) A=3: tan(B) A=7: tanh(B)Negatives produce the inverse of the respective functions | |||
| Deal | A distinct integers selected randomly from the first B integers | |||
| Membership, Epsilon | 1 for elements of A present in B; 0 where not. | |||
| Find, Epsilon Underbar | 1 for starting point of multi-item array A present in B; 0 where not. | |||
| Maximum, Ceiling | The greater value of A or B | |||
| Minimum, Floor | The smaller value of A or B | |||
| Reshape, Dyadic Rho | Array of shape A with data B | |||
| Take | Select the first (or last) A elements of B according to ×A | |||
| Drop | Remove the first (or last) A elements of B according to ×A | |||
| Decode | Value of a polynomial whose coefficients are B at A | |||
| Encode | Base-A representation of the value of B | |||
| Residue | B modulo A | |||
| Catenation | Elements of B appended to the elements of A | |||
| Expansion, Dyadic Backslash | Insert zeros (or blanks) in B corresponding to zeros in A | |||
| Compression, Dyadic Slash | Select elements in B corresponding to ones in A | |||
| Index of, Dyadic Iota | The location (index) of B in A; if not found | |||
| Solution to system of linear equations, multiple regression Ax = B | ||||
| Rotation | The elements of B are rotated A positions | |||
| Rotation | The elements of B are rotated A positions along the first axis | |||
| Logarithm | Logarithm of B to base A | |||
| Dyadic format | Format B into a character matrix according to A | |||
| General transpose | The axes of B are ordered by A | |||
| Combinations | Number of combinations of B taken A at a time | |||
| Diaeresis, Dieresis, Double-Dot | ||||
| Less than | Comparison: 1 if true, 0 if false | |||
| Less than or equal | Comparison: 1 if true, 0 if false | |||
| Equal | Comparison: 1 if true, 0 if false | |||
| Greater than or equal | Comparison: 1 if true, 0 if false | |||
| Greater than | Comparison: 1 if true, 0 if false | |||
| Not equal | Comparison: 1 if true, 0 if false | |||
| Or | Boolean Logic: 0 (False) if both A and B = 0, 1 otherwise. Alt: 1 (True) if A or B = 1 (True) | |||
| And | Boolean Logic: 1 (True) if both A and B = 1, 0 (False) otherwise | |||
| Nor | Boolean Logic: 1 if both A and B are 0, otherwise 0. Alt: ~∨ = not Or | |||
| Nand | Boolean Logic: 0 if both A and B are 1, otherwise 1. Alt: ~∧ = not And | |||
| Left | A | |||
| Right | B | |||
| Match | 0 if A does not match B exactly with respect to value, shape, and nesting; otherwise 1. | |||
| Laminate | Concatenate along first axis |
| Name(s) | Symbol | Example | Meaning (of example) | Unicode code point sequence | |
|---|---|---|---|---|---|
| Reduce (last axis), Slash | / | Sum across B | |||
| Reduce (first axis) | ⌿ | Sum down B | |||
| Scan (last axis), Backslash | \ | Running sum across B | |||
| Scan (first axis) | ⍀ | Running sum down B | |||
| Inner product | . | Matrix product of A and B | |||
| Outer product | ∘. | Outer product of A and B | , |
The product operator "." expects a dyadic function on both its left and right, forming a dyadic composite function applied to the vectors on its left and right. If the function to the left of the dot is "∘" (signifying null) then the composite function is an outer product, otherwise it is an inner product. An inner product intended for conventional matrix multiplication uses the + and × functions, replacing these with other dyadic functions can result in useful alternative operations.
Some functions can be followed by an axis indicator in (square) brackets, i.e., this appears between a function and an array and should not be confused with array subscripts written after an array. For example, given the ⌽ (reversal) function and a two-dimensional array, the function by default operates along the last axis but this can be changed using an axis indicator:
4 rows by 3 cols matrix created, using rho ⍴ and iota ⍳. The 4 x 3 matrix is then stored in a variable named A.
A is now reflected or flipped along its vertical axis as symbol ⌽ visually indicates.
A is now reflected using the [1] axis indicator or first dimension modifier. The result is that variable A has been reflected across the horizontal axis, instead of vertically.
A is now reflected both vertically ⊖ and horizontally ⌽.
A is ⍉ transposed to a 3 row by 4 col matrix such that rows-cols become exchanged, as symbol ⍉ visually portrays. Compare the result here to the original matrix stored in A, topmost matrix. These types of data transformations are useful in time series analysis and spatial coordinates, just two examples, more exist.
Arrays are structures which have elements grouped linearly as vectors or in table form as matrices—and higher dimensions (3D or cubed, 4D or cubed over time, etc.). Arrays containing both characters and numbers are termed mixed arrays.[9] Array structures containing elements which are also arrays are called nested arrays.[10]
| Explanation | ||
|---|---|---|
X set = to matrix with 4 rows by 5 columns, consisting of 20 consecutive integers. Element X[2;2] in row 2 - column 2 currently is an integer = 7. Initial index origin ⎕IO value = 1. Thus, the first element in matrix X or X[1;1] = 1. | ||
11 12 13 1 2 15 3 4 16 17 18 19 20 | Element in X[row 2; col 2] is changed (from 7) to a nested vector "Text" using the enclose ⊂ function. Element in X[row 3; col 4], formerly integer 14, now becomes a mini enclosed or ⊂ nested 2x2 matrix of 4 consecutive integers. Since X contains numbers, text and nested elements, it is both a mixed and a nested array. |
A user may define custom functions which, like variables, are identified by name rather than by a non-textual symbol. The function header defines whether a custom function is niladic (no arguments), monadic (one right argument) or dyadic (left and right arguments), the local name of the result (to the left of the ← assign arrow), and whether it has any local variables (each separated by semicolon ';').
Whether functions with the same identifier but different adicity are distinct is implementation-defined. If allowed, then a function CURVEAREA could be defined twice to replace both monadic CIRCLEAREA and dyadic SEGMENTAREA above, with the monadic or dyadic function being selected by the context in which it was referenced.Custom dyadic functions may usually be applied to parameters with the same conventions as built-in functions, i.e., arrays should either have the same number of elements or one of them should have a single element which is extended. There are exceptions to this, for example a function to convert pre-decimal UK currency to dollars would expect to take a parameter with precisely three elements representing pounds, shillings and pence.[11]
Inside a program or a custom function, control may be conditionally transferred to a statement identified by a line number or explicit label; if the target is 0 (zero) this terminates the program or returns to a function's caller. The most common form uses the APL compression function, as in the template (condition)/target which has the effect of evaluating the condition to 0 (false) or 1 (true) and then using that to mask the target (if the condition is false it is ignored, if true it is left alone so control is transferred).
Hence function SEGMENTAREA may be modified to abort (just below), returning zero if the parameters (DEGREES and RADIUS below) are of different sign:
The above function SEGMENTAREA works as expected if the parameters are scalars or single-element arrays, but not if they are multiple-element arrays since the condition ends up being based on a single element of the SIGN array - on the other hand, the user function could be modified to correctly handle vectorized arguments. Operation can sometimes be unpredictable since APL defines that computers with vector-processing capabilities should parallelise and may reorder array operations as far as possible - thus, test and debug user functions particularly if they will be used with vector or even matrix arguments. This affects not only explicit application of a custom function to arrays, but also its use anywhere that a dyadic function may reasonably be used such as in generation of a table of results:
A more concise way and sometimes better way - to formulate a function is to avoid explicit transfers of control, instead using expressions which evaluate correctly in all or the expected conditions. Sometimes it is correct to let a function fail when one or both input arguments are incorrect - precisely to let user know that one or both arguments used were incorrect. The following is more concise than the above SEGMENTAREA function. The below importantly correctly handles vectorized arguments:
90 180 270 ¯90 ∘.SEGMENTAREA 1 ¯2 40.785398163 0 12.56637061.57079633 0 25.13274122.35619449 0 37.69911180 ¯3.14159265 0
Avoiding explicit transfers of control also called branching, if not reviewed or carefully controlled - can promote use of excessively complex one liners, veritably "misunderstood and complex idioms" and a "write-only" style, which has done little to endear APL to influential commentators such as Edsger Dijkstra.[12] Conversely however APL idioms can be fun, educational and useful - if used with helpful comments ⍝, for example including source and intended meaning and function of the idiom(s). Here is an
APL idioms list, an IBM APL2 idioms list here[13] and Finnish APL idiom library here.| Name(s) | Symbol | Example | Meaning (of example) | Unicode code point | |
|---|---|---|---|---|---|
| High minus[14] | ¯ | Denotes a negative number | |||
| Lamp, Comment | ⍝ | Everything to the right of ⍝ denotes a comment | |||
| RightArrow, Branch, GoTo | → | →This_Label sends APL execution to This_Label: | |||
| Assign, LeftArrow, Set to | ← | B←A sets values and shape of B to match A |
Most APL implementations support a number of system variables and functions, usually preceded by the ⎕ (quad) and/or ")" (hook=close parenthesis) character. Note that the quad character is not the same as the Unicode missing character symbol. Particularly important and widely implemented is the ⎕IO (Index Origin) variable, since while the original IBM APL based its arrays on 1 some newer variants base them on zero:
| Description | ||
|---|---|---|
| X set = to vector of 12 consecutive integers. Initial index origin ⎕IO value = 1. Thus, the first position in vector X or X[1] = 1 per vector of iota values . | ||
| Index Origin ⎕IO now changed to 0. Thus, the 'first index position' in vector X changes from 1 to 0. Consequently, X[1] then references or points to 2 from and X[0] now references 1. | ||
| Quad WA or ⎕WA, another dynamic system variable, shows how much Work Area remains unused or 41,226 megabytes or about 41 gigabytes of unused additional total free work area available for the APL workspace and program to process using. If this number gets low or approaches zero - the computer may need more random-access memory (RAM), hard disk drive space or some combination of the two to increase virtual memory. | ||
| )VARS a system function in APL,[15] )VARS shows user variable names existing in the current workspace. |
There are also system functions available to users for saving the current workspace e.g., )SAVE and terminating the APL environment, e.g., )OFF - sometimes called hook commands or functions due to the use of a leading right parenthesis or hook.[16] There is some standardization of these quad and hook functions.
The Unicode Basic Multilingual Plane includes the APL symbols in the Miscellaneous Technical block,[17] which are thus usually rendered accurately from the larger Unicode fonts installed with most modern operating systems. These fonts are rarely designed by typographers familiar with APL glyphs. So, while accurate, the glyphs may look unfamiliar to APL programmers or be difficult to distinguish from one another.
Some Unicode fonts have been designed to display APL well: APLX Upright, APL385 Unicode, and SimPL.
Before Unicode, APL interpreters were supplied with fonts in which APL characters were mapped to less commonly used positions in the ASCII character sets, usually in the upper 128 code points. These mappings (and their national variations) were sometimes unique to each APL vendor's interpreter, which made the display of APL programs on the Web, in text files and manuals - frequently problematic.
Note the APL On/Off Key - topmost-rightmost key, just below. Also note the keyboard had some 55 unique (68 listed per tables above, including comparative symbols but several symbols appear in both monadic and dyadic tables) APL symbol keys (55 APL functions (operators) are listed in IBM's 5110 APL Reference Manual), thus with the use of alt, shift and ctrl keys - it would theoretically have allowed a maximum of some 59 (keys) *4 (with 2-key pressing) *3 (with tri-key pressing, e.g., ctrl-alt-del) or some 472 different maximum key combinations, approaching the 512 EBCDIC character max (256 chars times 2 codes for each keys-combination). Again, in theory the keyboard pictured here would have allowed for about 472 different APL symbols/functions to be keyboard-input, actively used. In practice, early versions were only using something roughly equivalent to 55 APL special symbols (excluding letters, numbers, punctuation, etc. keys). Thus, early APL was then only using about 11% (55/472) of a symbolic language's at-that-time utilization potential, based on keyboard # keys limits, again excluding numbers, letters, punctuation, etc. In another sense keyboard symbols utilization was closer to 100%, highly efficient, since EBCDIC only allowed 256 distinct chars, and ASCII only 128.
APL has proved to be extremely useful in solving mathematical puzzles, several of which are described below.
Take Pascal's triangle, which is a triangular array of numbers in which those at the ends of the rows are 1 and each of the other numbers is the sum of the nearest two numbers in the row just above it (the apex, 1, being at the top). The following is an APL one-liner function to visually depict Pascal's triangle:
Determine the number of prime numbers (prime # is a natural number greater than 1 that has no positive divisors other than 1 and itself) up to some number N. Ken Iverson is credited with the following one-liner APL solution to the problem:
Examining the converse or opposite of a mathematical solution is frequently needed (integer factors of a number): Prove for the subset of integers from 1 through 15 that they are non-prime by listing their decomposition factors. What are their non-one factors (#'s divisible by, except 1)?
ProveNonPrime ¨⍳15 ⍝ Prove non primes for each(¨) of the integers from 1 through 15 (iota 15) 1 prime 2 prime 3 prime 4 factors(except 1) 2 5 prime 6 factors(except 1) 2 3 7 prime 8 factors(except 1) 2 4 9 factors(except 1) 3 10 factors(except 1) 2 5 11 prime 12 factors(except 1) 2 3 4 6 13 prime 14 factors(except 1) 2 7 15 factors(except 1) 3 5
Generate a Fibonacci number sequence, where each subsequent number in the sequence is the sum of the prior two:
Fibonacci¨⍳14 ⍝ This APL statement says: Generate the Fibonacci sequence over each(¨) integer number(iota or ⍳) for the integers 1..14.0 1 1 2 3 5 8 13 21 34 55 89 144 233 ⍝ Generated sequence, i.e., the Fibonacci sequence of numbers generated by APL's interpreter.