The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
 / 
Tie-File-0.19
River stage three • 30 direct dependents • 100 total dependents
14 ++
/ Tie::File

NAME

Tie::File - Access the lines of a disk file via a Perl array

SYNOPSIS

        # This file documents Tie::File version 0.19

        tie @array, 'Tie::File', filename or die ...;

        $array[13] = 'blah';     # line 13 of the file is now 'blah'
        print $array[42];        # display line 42 of the file

        $n_recs = @array;        # how many records are in the file?
        $#array = $n_recs - 2;   # chop records off the end

        # As you would expect:

        push @array, new recs...;
        my $r1 = pop @array;
        unshift @array, new recs...;
        my $r1 = shift @array;
        @old_recs = splice @array, 3, 7, new recs...;

        untie @array;            # all finished

DESCRIPTION

Tie::File represents a regular text file as a Perl array. Each element in the array corresponds to a record in the file. The first line of the file is element 0 of the array; the second line is element 1, and so on.

The file is not loaded into memory, so this will work even for gigantic files.

Changes to the array are reflected in the file immediately.

Lazy people may now stop reading the manual.

recsep

What is a 'record'? By default, the meaning is the same as for the <...> operator: It's a string terminated by $/, which is probably "\n". (Minor exception: on dos and Win32 systems, a 'record' is a string terminated by "\r\n".) You may change the definition of "record" by supplying the recsep option in the tie call:

        tie @array, 'Tie::File', $file, recsep => 'es';

This says that records are delimited by the string es. If the file contained the following data:

        Curse these pesky flies!\n

then the @array would appear to have four elements:

        "Curse thes"
        "e pes"
        "ky flies"
        "!\n"

An undefined value is not permitted as a record separator. Perl's special "paragraph mode" semantics (à la $/ = "") are not emulated.

Records read from the tied array will have the record separator string on the end, just as if they were read from the <...> operator. Records stored into the array will have the record separator string appended before they are written to the file, if they don't have one already. For example, if the record separator string is "\n", then the following two lines do exactly the same thing:

        $array[17] = "Cherry pie";
        $array[17] = "Cherry pie\n";

The result is that the contents of line 17 of the file will be replaced with "Cherry pie"; a newline character will separate line 17 from line 18. This means that in particular, this will do nothing:

        chomp $array[17];

Because the chomped value will have the separator reattached when it is written back to the file. There is no way to create a file whose trailing record separator string is missing.

Inserting records that contain the record separator string will produce a reasonable result, but if you can't foresee what this result will be, you'd better avoid doing this.

mode

Normally, the specified file will be opened for read and write access, and will be created if it does not exist. (That is, the flags O_RDWR | O_CREAT are supplied in the open call.) If you want to change this, you may supply alternative flags in the mode option. See Fcntl for a listing of available flags. For example:

        # open the file if it exists, but fail if it does not exist
        use Fcntl 'O_RDWR';
        tie @array, 'Tie::File', $file, mode => O_RDWR;

        # create the file if it does not exist
        use Fcntl 'O_RDWR', 'O_CREAT';
        tie @array, 'Tie::File', $file, mode => O_RDWR | O_CREAT;

        # open an existing file in read-only mode
        use Fcntl 'O_RDONLY';
        tie @array, 'Tie::File', $file, mode => O_RDONLY;

Opening the data file in write-only or append mode is not supported.

memory

This is an (inexact) upper limit on the amount of memory that Tie::File will consume at any time while managing the file. At present, this is used as a bound on the size of the read cache.

Records read in from the file are cached, to avoid having to re-read them repeatedly. If you read the same record twice, the first time it will be stored in memory, and the second time it will be fetched from the read cache. The amount of data in the read cache will not exceed the value you specified for memory. If Tie::File wants to cache a new record, but the read cache is full, it will make room by expiring the least-recently visited records from the read cache.

The default memory limit is 2Mib. You can adjust the maximum read cache size by supplying the memory option. The argument is the desired cache size, in bytes.

        # I have a lot of memory, so use a large cache to speed up access
        tie @array, 'Tie::File', $file, memory => 20_000_000;

Setting the memory limit to 0 will inhibit caching; records will be fetched from disk every time you examine them.

Option Format

-mode is a synonym for mode. -recsep is a synonym for recsep. -memory is a synonym for memory. You get the idea.

Public Methods

The tie call returns an object, say $o. You may call

        $rec = $o->FETCH($n);
        $o->STORE($n, $rec);

to fetch or store the record at line $n, respectively; similarly the other tied array methods. (See perltie for details.) You may also call the following methods on this object:

flock

        $o->flock(MODE)

will lock the tied file. MODE has the same meaning as the second argument to the Perl built-in flock function; for example LOCK_SH or LOCK_EX | LOCK_NB. (These constants are provided by the use Fcntl ':flock' declaration.)

MODE is optional; $o->flock simply locks the file with LOCK_EX.

The best way to unlock a file is to discard the object and untie the array. It is probably unsafe to unlock the file without also untying it, because if you do, changes may remain unwritten inside the object. That is why there is no shortcut for unlocking. If you really want to unlock the file prematurely, you know what to do; if you don't know what to do, then don't do it.

All the usual warnings about file locking apply here. In particular, note that file locking in Perl is advisory, which means that holding a lock will not prevent anyone else from reading, writing, or erasing the file; it only prevents them from getting another lock at the same time. Locks are analogous to green traffic lights: If you have a green light, that does not prevent the idiot coming the other way from plowing into you sideways; it merely guarantees to you that the idiot does not also have a green light at the same time.

Tying to an already-opened filehandle

If $fh is a filehandle, such as is returned by IO::File or one of the other IO modules, you may use:

        tie @array, 'Tie::File', $fh, ...;

Similarly if you opened that handle FH with regular open or sysopen, you may use:

        tie @array, 'Tie::File', \*FH, ...;

Handles that were opened write-only won't work. Handles that were opened read-only will work as long as you don't try to write to them. Handles must be attached to seekable sources of data---that means no pipes or sockets. If you supply a non-seekable handle, the tie call will try to abort your program.

CAVEATS

(That's Latin for 'warnings'.)

  • This is BETA RELEASE SOFTWARE. It may have bugs. See the discussion below about the (lack of any) warranty.

  • Every effort was made to make this module efficient. Nevertheless, changing the size of a record in the middle of a large file will always be fairly slow, because everything after the new record must be moved.

    In particular, note that the following innocent-looking loop has very bad behavior:

            # million-line file
        for (@file_array) {
          $_ .= 'x';
        }

    This is likely to be very slow, because the first iteration must relocate lines 1 through 999,999; the second iteration must relocate lines 2 through 999,999, and so on. The relocation is done using block writes, however, so it's not as slow as it might be.

    A soon-to-be-released version of this module will provide a mechanism for getting better performance in such cases, by deferring the writing until it can be done all at once. This deferred writing feature might be enabled automagically if Tie::File guesses that you are about to write many consecutive records. To disable this feature, use

            (tied @o)->autodefer(0);

    (At present, this call does nothing.)

  • The behavior of tied arrays is not precisely the same as for regular arrays. For example:

            undef $a[10];  print "How unusual!\n" if $a[10];

    undef-ing a Tie::File array element just blanks out the corresponding record in the file. When you read it back again, you'll see the record separator (typically, $a[10] will appear to contain "\n") so the supposedly-undef'ed value will be true.

    There are other minor differences, but in general, the correspondence is extremely close.

  • Not quite every effort was made to make this module as efficient as possible. FETCHSIZE should use binary search instead of linear search. The cache's LRU queue should be a heap instead of a list. These defects are probably minor; in any event, they will be fixed in a later version of the module.

  • The author has supposed that since this module is concerned with file I/O, almost all normal use of it will be heavily I/O bound, and that the time to maintain complicated data structures inside the module will be dominated by the time to actually perform the I/O. This suggests, for example, that an LRU read-cache is a good tradeoff, even if it requires substantial adjustment following a splice operation.

WHAT ABOUT DB_File?

DB_File's DB_RECNO feature does something similar to Tie::File, but there are a number of reasons that you might prefer Tie::File. DB_File is a great piece of software, but the DB_RECNO part is less great than the rest of it.

  • DB_File reads your entire file into memory, modifies it in memory, and the writes out the entire file again when you untie the file. This is completely impractical for large files.

    Tie::File does not do any of those things. It doesn't try to read the entire file into memory; instead it uses a lazy approach and caches recently-used records. The cache size is strictly bounded by the memory option. DB_File's ->{cachesize} doesn't prevent your process from blowing up when reading a big file.

  • DB_File has an extremely poor writing strategy. If you have a ten-megabyte file and tie it with DB_File, and then use

            $a[0] =~ s/PERL/Perl/;

    DB_file will then read the entire ten-megabyte file into memory, do the change, and write the entire file back to disk, reading ten megabytes and writing ten megabytes. Tie::File will read and write only the first record.

    If you have a million-record file and tie it with DB_File, and then use

            $a[999998] =~ s/Larry/Larry Wall/;

    DB_File will read the entire million-record file into memory, do the change, and write the entire file back to disk. Tie::File will only rewrite records 999998 and 999999. During the writing process, it will never have more than a few kilobytes of data in memory at any time, even if the two records are very large.

  • Since changes to DB_File files only appear when you do untie, it can be inconvenient to arrange for concurrent access to the same file by two or more processes. Each process needs to call $db->sync after every write. When you change a Tie::File array, the changes are reflected in the file immediately; no explicit ->sync call is required. (The forthcoming "deferred writing" mode will allow you to request that writes be held in memory until explicitly sync'ed.)

  • DB_File is only installed by default if you already have the db library on your system; Tie::File is pure Perl and is installed by default no matter what. Starting with Perl 5.7.3 you can be absolutely sure it will be everywhere. You will never have that surety with DB_File. If you don't have DB_File yet, it requires a C compiler. You can install Tie::File from CPAN in five minutes with no compiler.

  • DB_File is written in C, so if you aren't allowed to install modules on your system, it is useless. Tie::File is written in Perl, so even if you aren't allowed to install modules, you can look into the source code, see how it works, and copy the subroutines or the ideas from the subroutines directly into your own Perl program.

  • Except in very old, unsupported versions, DB_File's free license requires that you distribute the source code for your entire application. If you are not able to distribute the source code for your application, you must negotiate an alternative license from Sleepycat, possibly for a fee. Tie::File is under the Perl Artistic license and can be distributed free under the same terms as Perl itself.

AUTHOR

Mark Jason Dominus

To contact the author, send email to: mjd-perl-tiefile+@plover.com

To receive an announcement whenever a new version of this module is released, send a blank email message to mjd-perl-tiefile-subscribe@plover.com.

LICENSE

Tie::File version 0.19 is copyright (C) 2002 Mark Jason Dominus.

This library is free software; you may redistribute it and/or modify it under the same terms as Perl itself.

These terms include your choice of (1) the Perl Artistic Licence, or (2) version 2 of the GNU General Public License as published by the Free Software Foundation, or (3) any later version of the GNU General Public License.

This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this library program; it should be in the file COPYING. If not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111 USA

For licensing inquiries, contact the author at:

        Mark Jason Dominus
        255 S. Warnock St.
        Philadelphia, PA 19107

WARRANTY

Tie::File version 0.19 comes with ABSOLUTELY NO WARRANTY. For details, see the license.

THANKS

Gigantic thanks to Jarkko Hietaniemi, for agreeing to put this in the core when I hadn't written it yet, and for generally being helpful, supportive, and competent. (Usually the rule is "choose any one.") Also big thanks to Abhijit Menon-Sen for all of the same things.

Special thanks to Craig Berry (for VMS portability help), Randy Kobes (for Win32 portability help), Clinton Pierce and Autrijus Tang (for heroic eleventh-hour Win32 testing above and beyond the call of duty), and the rest of the CPAN testers (for testing generally).

More thanks to: Edward Avis / Gerrit Haase / Nikola Knezevic / Nick Ing-Simmons / Tassilo von Parseval / H. Dieter Pearcey / Slaven Rezic / Peter Somu / Tels

TODO

Test DELETE machinery more carefully.

More tests. (mode option. _twrite should be tested separately, because there are a lot of weird special cases lurking in there.)

More tests. (Stuff I didn't think of yet.)

Paragraph mode?

More tests.

Fixed-length mode.

Maybe an autolocking mode?

Finish deferred writing.

Autodeferment.

Record locking with fcntl()? Then you might support an undo log and get real transactions. What a coup that would be.

Leave-blanks mode