NAME

Tie::MmapArray - mmap a file as a tied array

SYNOPSIS

    use Tie::MmapArray;

    tie @array, 'Tie::MmapArray', $filename;
    tie @array, 'Tie::MmapArray', $filename, $template;
    tie @array, 'Tie::MmapArray', $filename, { template => $template,
                                               nels     => 0,
                                               mode     => "rw",
                                               shared   => 1,
                                               offset   => 0 };

DESCRIPTION

The Tie::MmapArray module lets you use mmap to map in a file as a perl array rather than reading the file into dynamically allocated memory. It depends on your operating system supporting UNIX or POSIX.1b mmap, of course. (Code to use the equivalent functions on Win32 platforms has been contributed but has not been tested yet.)

The type of array elements is defined by the template argument or option. This is a Perl pack()-style template, which defaults to ``i''. The template may be an array reference, in which case the elements are defined by pairs of name and template for each element. A template string may define multiple fields, in which case that element is regarded as an array of fields (which need not be of the same type).

The following example shows the utmp file on Linux mapped to an array:

    tie @utmp, 'Tie::MmapArray', '/var/log/utmp',
        [
          ut_type    => 's',
          ut_pid     => 'i',    # pid_t
          ut_line    => 'a12',
          ut_id      => 'a4',
          ut_user    => 'a32',
          ut_host    => 'a256',
          ut_exit    => [ # struct exit_status
                          e_termination => 's',
                          e_exit        => 's' ],
          ut_session => 'l',
          ut_tv      => [ # struct timeval
                          tv_sec  => 'l'
                          tv_usec => 'l' ],
          ut_addr_v6 => 'l4',
          pad        => 'a20' ];

This can be scanned as follows:

    for (my $i = 0; $i < @utmp; $i++) {
        print("pid: %d, user: %s\n",
              $utmp[$i]->{ut_pid}, $utmp[$i]->{ut_user});
    }

The following subset of pack() template letters is supported:

i

signed integer (default)

I

unsigned integer

c

signed character (one byte integer)

c

unsigned character (one byte integer)

s

signed short integer

S

unsigned short integer

n

unsigned short integer in network byte order

l

signed long integer

L

unsigned long integer

N

unsigned long integer in network byte order

f

float

d

double

aN

fixed-length, null-padded ASCII string of length N

AN

fixed-length, space-padded ASCII string of length N

The size of the array is defined by the nels option. If this is zero then it is calculated as the file size divided by the element size.

If the file size is smaller than the size required for the requested elements then a single zero byte will be written to the final byte of the requested size. This seems to prevent the module dying with a segmentation or bus error if memory is accessed beyond the end of the file and generally results in a file with holes (unallocated blocks). Precise details of the behaviour of the module are subject to change.

BUGS, RESTRICTIONS AND FUTURE DIRECTIONS

This is version 0.02 of the module and there are likely to be many bugs. The interface may change as the result of feedback.

The options mode and shared are not yet used.

Not all pack letters are implemented yet.

push, pop, shift, unshift, and splice operations are not yet supported. It is debateable whether they should be as they could be very expensive if the mmaped file was large (say a Gigabyte or two). Perhaps there should be an option to explicitly allow these operations.

AUTHOR

Andrew Ford <[email protected]>, 27 December 1999.

CREDITS

The module was inspired by Malcolm Beatie's Mmap module.

Reini Urban <[email protected]> provided intial code for Win32 platforms.