Comparison of APL dialects

All APL dialects share core features, and at least the primitives are backwards compatible with those of APL.SV. However, there are important dialectal differences, the most significant of which are described in this article.

Flat versus Nested
One of the most significant differences between APLs is the choice of array model, namely nested or flat. Originally, all APLs were entirely flat, that is, the only scalar values supported were simple and had to be either numbers and characters, and an array had to be homogeneous, that is, consisting entirely of a single type. SHARP APL introduced the box as a third type, such that a box could hold any other array (including a box array), and this allowed for arrays contain other arrays. NARS introduced the nested array model, where any arrays can be an actual element of another array. Most currently maintained dialects (Dyalog APL, NARS2000, APL2, GNU APL, and others) use the nested array model.

Disclose
Disclose is defined, on scalars, to be the inverse of Enclose, giving the sole element of that scalar. This behavior is extended to arbitrary arrays in one of two ways: Mix, which combines all elements, and First, which returns only the first (in ravel order).

In the SHARP APL family, Disclose is written, and given a function rank of 0, which causes it to behave as Mix. First is not a primitive.

In nested APL dialects, Disclose is written, and it was extended to First by NARS and to Mix by APL2. The other possible meaning is given the glyph  and named "Mix" or "First" (not "Disclose") as appropriate. Dialects might choose either assignment of glyphs, and Dyalog APL and APL*PLUS allow either to be chosen based on Migration Level.

Numeric types
Traditionally, APL supports only one numeric type that is visible to the user, such as a floating-point number, or complex number whose components are floating-point. Other numeric types may be used by the implementation, but only for optimization: for example, APL\360 uses packed 1-bit Boolean and 4-bit integer arrays that are automatically converted to float if an operation yields a value that doesn't fit in the type. ISO/IEC 13751:2001 specifies that different representations of a number must be indistinguishable to a conforming program (which can call primitives but not implementation-specific system functions).

Most often the numeric type is a 64-bit float, for instance HFP in early IBM dialects and IEEE 754 in new dialects since the 1980s. Dyalog APL supports 128-bit decimal floats by setting. Modern dialects tend to support complex numbers, with exceptions including dzaima/APL, APLX, and APL+Win. Dialects with complex numbers also extend Circular to left arguments designed for complex numbers.

Some newer dialects provide additional types that are distinguishable from floats, often by greater precision. Extended-precision integers are one possibility: for example, dzaima/APL provides these if requested with a trailing  in a numeric literal or , while Kap uses exact rational numbers by default. NARS2000 supports numerous numeric types, and April provides access to the Common Lisp numeric tower, which also includes many types. GNU APL doesn't support arbitrary precision, but uses 64-bit integers by default, much like J or K.

Conformability
All APLs implement scalar extension in scalar dyadic functions, and Each if present. However, a few dialects differ in their support for singleton extension. Most support it fully, so that a singleton of any rank is extended. APL2 supports it only for scalars and one-element vectors (and ISO/IEC_13751:2001 specifies only this behavior), and dzaima/APL and Kap support scalar extension only.

Furthermore, dialects that implement the Rank operator differ in their conformability rules for frames. Dyalog APL and SHARP APL implement scalar agreement only, NARS2000 implements singleton agreement, and A+ implements leading axis agreement (in addition to extra options allowing the programmer to specify outer-product-like treatment of some axes).

Index origin
A few dialects fix the index origin to 0 rather than allowing it to be configured: A+, APL, ngn/apl, Co-dfns, and Kap.

Depth
Dyalog APL defines Depth to return negative numbers when the depth is uneven. Most other dialects instead define it to return the maximum depth. The original NARS paper assigned the glyph to a function called Simple, which simply answered if the argument was simple or not, without reporting the actual depth. It was thus equivalent to today's.

Language features
The following table shows support for various APL features: nesting either with boxes or the nested array model, defined functions or similar syntax, control structures, dfn-style anonymous functions, tacit programming, particularly with trains, complex numbers, arbitrary-precision integers, Unicode support, and namespaces or a similar encapsulation feature. The precise syntax and capabilities of these features might differ between dialects that support them.

Dialects included in the first row include APL\360, APL\1130, APL*PLUS before II, APL/700, APL.SV, APL 5100, MCM/70, VS APL, APL\3000, and Aplette. Extended Dyalog APL and Dyalog APL Vision have the same feature-set as Dyalog APL.

Development and platform
Shown are the release year, year of last release for inactive languages, withdrawal year for discontinued products, developer, implementation language, and host system. Systems capable of being compiled for multiple native hosts are indicated as "portable".