This and other RFCs are available on the web at http://dev.perl.org/rfc/ =head1 TITLE Arrays: Notation for declaring and creating arrays =head1 VERSION Maintainer: Jeremy Howard <[EMAIL PROTECTED]> Date: 8 Sep 2000 Last Modified: 21 Sep 2000 Mailing List: [EMAIL PROTECTED] Number: 203 Version: 2 Status: Frozen =head1 DISCUSSION No objections were noted to the proposals in this RFC. A change of name from :bounds to :shape was accepted. =head1 ABSTRACT RFC 202 described the need to be able to declare a data structure that contains elements of the same type stored contiguously in memory, which is called an I<array>. This RFC outlines the syntax to declare and create arrays. The syntax to create arrays is identical to that to create lists of lists (described in L<perllol> in the Perl 5 documentation). The syntax to declare the type of elements is the standard type syntax. RFC 203 describes a syntax for multidimensional indexing of arrays. A syntax to declare the bounds of the dimensions is described in this RFC using a new ':shape' attribute. =head1 DESCRIPTION =head2 Compact arrays It is proposed that if a list is declared that specifies a simple type for its elements: my int @a; then that list be stored as an array--that is, contiguously in memory. These arrays support I<all> the same syntax as lists. Therefore any description of syntax for a 'list' also applies to an 'array', and visa versa. However, their implementation is very different. =head2 :shape attribute Furthermore, it is proposed that lists accept a new C<:shape> attribute: my @a :shape(3); that defines the number of elements in a list. This is equivalent to: my @a; $#a = 3-1; except that an attempt at accessing beyond $a[3] would result in an error if :shape(3) is set. Specfically, adding a :shape attribute to a declaration has two effects on the array: =over 4 =item 1 Adds range checking (since exceptions are defined for access outside the range) =item 2 Allows the block of memory to be preallocated =back (2) can also be achieved by simply setting the bottom right element to undef, or by setting @#array. The behaviour of (1) removes autovivification of new elements, since an exception is raised instead. :shape doesn't actually reshape. If the returned array overshoots specified bounds of :shape, an exception is raised. Reshaping is done with reshape() (RFC 148), merge()/demerge() (RFC 90), and part()/flatten() (RFC 91). The :shape attribute can also accept a list: my @b :shape(3,3); The second element of the list is the number of elements in the list, as before. The first element is the number of lists that are referenced as elements of the list. Therefore my @b :shape(3,3); $b[3][4] = 0; # Error: access beyond bounds of @b would result in an error. :shape can take as many arguments as required--an n-element list declares a list with at most n levels of nesting, with the maximum index at level x being (n-x). Because lists of lists support multidimensional indexing (see RFC 204) the :shape attribute effectively specifies the bounds of a multidimensional structure. The parameters of :shape are optional if an array is assigned in the declaration. Therefore: my int @array :shape = @rvalue; is equivalent to: my int @array :shape(@#rvalue) = @rvalue; The bounds of an array or list can be specified at run time, of course: my @t1 :shape(@dimList) = getFromSomeplace(); =head2 Combining compact storage and :shape attribute Efficient multidimensional arrays can be declared by combining a fixed simple type with the :shape attribute: my int @b :shape(4,4); Perl in this case would set aside enough room for sixteen ints, and store an attribute with @b that it had two dimensions, each indexed by (0..3). Because @b here is stored as an array, and supports multidimensional indexing (see RFC 204), it is a true multidimensional array. Although @b looks just like a normal list of lists that happens to have a type and an attribute, it is implemented as a multidimensional array. Therefore my int @b :shape(4,4); @b = ([1,2,3,4], [5,6,7,8], [9,10,11,12], [13,14,15,16]); creates a multidimensional array @b that contains all sixteen ints in a contiguous block of memory, but can be accessed using standard list of lists syntax, along with the extensions proposed in RFC 204. Where the type and bounds of an array can be derived at run time, it is not necessary to specify them explicitly: my int @t1 :shape(@dimList) = getFromSomeplace(); my int @t2 :shape(@dimList) = getFromSomeplaceElse(); my @prod = @t1 * @t2; # @prod magically has type (int) and :shape (@dimlist) Note that this is using an element-wise multiplication operation, described in RFC 82. If either @t1 or @t2 was unbounded (i.e. had no :shape attribute) then @prod would also be unbounded. A list (of lists...) that contains elements of the same type can be converted to an array by specifying its type: my @some_LOL = ([1,2], [3,4]); my int @array = @some_LOL; and its bounds can be locked in as well if required: my @some_LOL = ([1,2], [3,4]); my int @array :shape(@#some_LOL) = @some_LOL; =head1 IMPLEMENTATION Too early to get into much detail beyond the obvious... Clearly arrays are not a list of SVs, but are the raw data stored contiguously in memory, along with the attributes of the array stored someplace. :shape applies to lists as well, of course (because lists and arrays share identical syntax), but this should be fine because it is just an attribute. Arrays do not require :shape to be specified. If not specified, they should grow by doubling in size (like lists), but programmers will be encouraged to avoid this because a new contiguous memory block will have to be found each time. Programmers may also allocate memory after the fact with '@#' (see RFC 206). =head1 REFERENCES RFC 202: Overview of multidimensional array RFCs perllol in the Perl 5 documentation Implementation in PDL: http://pdl.sourceforge.net/PDLdocs/Internals.html