This and other RFCs are available on the web at http://dev.perl.org/rfc/ =head1 TITLE Array: Use list reference for multidimensional array access =head1 VERSION Maintainer: Buddha Buck <[EMAIL PROTECTED]> Date: 8 September 2000 Mailing List: [EMAIL PROTECTED] Number: 204 Version: 1 Status: Developing =head1 ABSTRACT This RFC proposes that array indexing syntax be extended to use references to integer lists as indices for multidimensional arrays. =head1 DESCRIPTION Currently in Perl5, arrays can be indexed by a single integer, or sliced by a list of integers. This notation is sufficient and convenient for one dimensional arrays (like Perl5 has), but is problematic for multidimensional arrays. Because n-dimensional arrays have more than one dimension, their indices are not simple integers, but rather n-tuples of integers. While lists are natural Perl representations of n-tuples, they present problems when used as multidimensional array indices: =over 4 =item * It is ambiguous if @a[@point1,@point2] is a 2-element slice of a multidimensional array, or if it is a multi-element slice of a single dimensional array. =item * Although $a[@point] is not as ambiguous (it is currently a syntax error), it still could be confusing, especially when seeing both $a[1,3,4] and @b[1,2,4]. =item * It is inconvenient use a list of n-dimensional indices for a slice. By Perl syntax, such a list would have to be a list of lists, whereas currently Perl only allows lists of scalers. =back A solution to these problems is to encapsulate a list of integers into a scaler, such as using a list reference. By using list references for indices (and lists of list references for slices), the multidimensional case is similar to the unidimensional case: $a[$scaler] would access a single element, @a[@list] would access a slice. This would solve the problems above: =over 4 =item * @a[$point1,$point2] is no longer ambiguous, but obviously a 2-element slice from the array @a. The dimensionality of @a is not an issue with this syntax. =item * Multidimensional and unidimensional array access both use $a[$point], so $a[1,3,4] will not be seen. Instead, any confusion would have to be between $a[[1,3,4]] and @b[1,2,4]. Since $a[[1,3,4]] is an array indexed by a single scaler value (the listref [1,3,4]), whereas @b[1,2,4] is an array indexed by a list of scaler values explains clearly (and possibly redundantly) the use of $ and @ as prefixes. =item * List of lists are naturally stored as list of scalar references to lists. As such, the natural representation of lists of lists ties in well with taking multidimensional array slices: for my $t (1..10) { push @points,[getx($t),gety($t),getz($t)]; } $point = $points[3]; @array[@points] = 1 x @points; $array[$point] = 0; =back This proposed syntax extends to unidimensional arrays as well. $a[[$b]] being an alternative way of saying $a[$b], when $b is an integer. $a[[]] would return the (unique) value of the zero-dimensional array @a, not the value of the scaler $a. A listref used as an index must have the same or lower cardinality as the dimensionality of the array (if the array is declared with ':bounds'-- see RFC 203), and must have elements within the declared or implied bounds of the array indexed. my int @array :bounds(3,3,3); $array[[0,1,2]] = 5; # OK, within bounds. $array[[0,1,4]] = 5; # error, out of bounds $array[[1,2,1,2]] = 5; # error, out of bounds, too many dimensions If the cardinality (c) of the index is lower than the dimensions (d) of the array, then the innermost (n-d) dimensions of the array are returned: my int @array :bounds(3,3,3); $array[[1,2]] = (1,2,3,4); # Sets the line at (1,2) to (1,2,3,4) When a listref is used to index a list of lists, the returned list reference is automatically dereferenced: my @array( [0,1], [1,2]); my @a = @array[[0]]; # Returns (0,1), _not_ [0,1] The ; operator (defined in RFC 205) is designed to return lists of list refs, to make for efficient indexing and slicing of multidimensional arrays. @points = ((1,2);(3,4);5); @points = ([1,3,5],[1,4,5],[2,3,5],[2,4,5]) @array[@points]= 5 x @points; =head1 IMPLEMENTATION Implementation of this feature is dependent on the underlying implementation of multidimensional arrays. Other than that, it is a straightforward enhancement of the semantics of list indexing. =head1 REFERENCES RFC 202: Overview of multidimensional array RFCs RFC 203: Notation for declaring and creating arrays RFC 205: New operator ';' for creating array slices. =head1 ACKNOWLEDGEMENTS Jeremy Howard: Suggested listref indexing
