package DBM::Deep::Engine;
+use 5.6.0;
+
use strict;
+use warnings;
use Fcntl qw( :DEFAULT :flock :seek );
+use Scalar::Util ();
+
+# File-wide notes:
+# * All the local($/,$\); are to protect read() and print() from -l.
+# * To add to bucket_size, make sure you modify the following:
+# - calculate_sizes()
+# - _get_key_subloc()
+# - add_bucket() - where the buckets are printed
##
# Setup file and tag signatures. These should never change.
##
sub SIG_FILE () { 'DPDB' }
+sub SIG_HEADER () { 'h' }
sub SIG_INTERNAL () { 'i' }
sub SIG_HASH () { 'H' }
sub SIG_ARRAY () { 'A' }
-sub SIG_SCALAR () { 'S' }
sub SIG_NULL () { 'N' }
sub SIG_DATA () { 'D' }
sub SIG_INDEX () { 'I' }
sub SIG_BLIST () { 'B' }
+sub SIG_FREE () { 'F' }
sub SIG_SIZE () { 1 }
-sub precalc_sizes {
- ##
- # Precalculate index, bucket and bucket list sizes
- ##
- my $self = shift;
-
- $self->{index_size} = (2**8) * $self->{long_size};
- $self->{bucket_size} = $self->{hash_size} + $self->{long_size};
- $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
-
- return 1;
-}
-
-sub set_pack {
- ##
- # Set pack/unpack modes (see file header for more)
- ##
- my $self = shift;
- my ($long_s, $long_p, $data_s, $data_p) = @_;
-
- ##
- # Set to 4 and 'N' for 32-bit offset tags (default). Theoretical limit of 4
- # GB per file.
- # (Perl must be compiled with largefile support for files > 2 GB)
- #
- # Set to 8 and 'Q' for 64-bit offsets. Theoretical limit of 16 XB per file.
- # (Perl must be compiled with largefile and 64-bit long support)
- ##
- $self->{long_size} = $long_s ? $long_s : 4;
- $self->{long_pack} = $long_p ? $long_p : 'N';
-
- ##
- # Set to 4 and 'N' for 32-bit data length prefixes. Limit of 4 GB for each
- # key/value. Upgrading this is possible (see above) but probably not necessary.
- # If you need more than 4 GB for a single key or value, this module is really
- # not for you :-)
- ##
- $self->{data_size} = $data_s ? $data_s : 4;
- $self->{data_pack} = $data_p ? $data_p : 'N';
-
- return $self->precalc_sizes();
-}
-
-sub set_digest {
- ##
- # Set key digest function (default is MD5)
- ##
- my $self = shift;
- my ($digest_func, $hash_size) = @_;
-
- $self->{digest} = $digest_func ? $digest_func : \&Digest::MD5::md5;
- $self->{hash_size} = $hash_size ? $hash_size : 16;
-
- return $self->precalc_sizes();
-}
-
sub new {
my $class = shift;
my ($args) = @_;
hash_size => 16,
##
- # Maximum number of buckets per list before another level of indexing is done.
- # Increase this value for slightly greater speed, but larger database files.
- # DO NOT decrease this value below 16, due to risk of recursive reindex overrun.
+ # Maximum number of buckets per list before another level of indexing is
+ # done. Increase this value for slightly greater speed, but larger database
+ # files. DO NOT decrease this value below 16, due to risk of recursive
+ # reindex overrun.
##
max_buckets => 16,
+
+ fileobj => undef,
+ obj => undef,
}, $class;
- $self->precalc_sizes;
+ if ( defined $args->{pack_size} ) {
+ if ( lc $args->{pack_size} eq 'small' ) {
+ $args->{long_size} = 2;
+ $args->{long_pack} = 'n';
+ }
+ elsif ( lc $args->{pack_size} eq 'medium' ) {
+ $args->{long_size} = 4;
+ $args->{long_pack} = 'N';
+ }
+ elsif ( lc $args->{pack_size} eq 'large' ) {
+ $args->{long_size} = 8;
+ $args->{long_pack} = 'Q';
+ }
+ else {
+ die "Unknown pack_size value: '$args->{pack_size}'\n";
+ }
+ }
+
+ # Grab the parameters we want to use
+ foreach my $param ( keys %$self ) {
+ next unless exists $args->{$param};
+ $self->{$param} = $args->{$param};
+ }
+ Scalar::Util::weaken( $self->{obj} ) if $self->{obj};
+
+ if ( $self->{max_buckets} < 16 ) {
+ warn "Floor of max_buckets is 16. Setting it to 16 from '$self->{max_buckets}'\n";
+ $self->{max_buckets} = 16;
+ }
return $self;
}
-sub setup_fh {
- my $self = shift;
- my ($obj) = @_;
+sub _fileobj { return $_[0]{fileobj} }
+sub _fh { return $_[0]->_fileobj->{fh} }
- $self->open( $obj ) if !defined $obj->_fh;
+sub calculate_sizes {
+ my $self = shift;
- #XXX We have to make sure we don't mess up when autoflush isn't turned on
- unless ( $obj->_root->{inode} ) {
- my @stats = stat($obj->_fh);
- $obj->_root->{inode} = $stats[1];
- $obj->_root->{end} = $stats[7];
- }
+ #XXX Does this need to be updated with different hashing algorithms?
+ $self->{index_size} = (2**8) * $self->{long_size};
+ $self->{bucket_size} = $self->{hash_size} + $self->{long_size} * 3;
+ $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
- return 1;
+ return;
}
-sub open {
- ##
- # Open a fh to the database, create if nonexistent.
- # Make sure file signature matches DBM::Deep spec.
- ##
+sub write_file_header {
my $self = shift;
- my ($obj) = @_;
-
- if (defined($obj->_fh)) { $self->close_fh( $obj ); }
-
- # Theoretically, adding O_BINARY should remove the need for the binmode
- # Of course, testing it is going to be ... interesting.
- my $flags = O_RDWR | O_CREAT | O_BINARY;
- my $fh;
- sysopen( $fh, $obj->_root->{file}, $flags )
- or $obj->_throw_error("Cannot sysopen file: " . $obj->_root->{file} . ": $!");
- $obj->_root->{fh} = $fh;
+ local($/,$\);
+
+ my $fh = $self->_fh;
+
+ my $loc = $self->_request_space( length( SIG_FILE ) + 21 );
+ seek($fh, $loc + $self->_fileobj->{file_offset}, SEEK_SET);
+ print( $fh
+ SIG_FILE,
+ SIG_HEADER,
+ pack('N', 1), # header version
+ pack('N', 12), # header size
+ pack('N', 0), # currently running transaction IDs
+ pack('n', $self->{long_size}),
+ pack('A', $self->{long_pack}),
+ pack('n', $self->{data_size}),
+ pack('A', $self->{data_pack}),
+ pack('n', $self->{max_buckets}),
+ );
- #XXX Can we remove this by using the right sysopen() flags?
- # Maybe ... q.v. above
- binmode $fh; # for win32
+ $self->_fileobj->set_transaction_offset( 13 );
- if ($obj->_root->{autoflush}) {
- my $old = select $fh;
- $|=1;
- select $old;
- }
+ return;
+}
- seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
+sub read_file_header {
+ my $self = shift;
- my $signature;
- my $bytes_read = read( $fh, $signature, length(SIG_FILE));
+ local($/,$\);
- ##
- # File is empty -- write signature and master index
- ##
- if (!$bytes_read) {
- seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET);
- print( $fh SIG_FILE);
+ my $fh = $self->_fh;
- $self->create_tag($obj, $obj->_base_offset, $obj->_type, chr(0) x $self->{index_size});
+ seek($fh, 0 + $self->_fileobj->{file_offset}, SEEK_SET);
+ my $buffer;
+ my $bytes_read = read( $fh, $buffer, length(SIG_FILE) + 9 );
- # Why is this being printed here? I'm not seeing where anything actually points to
- # this spot.
- #XXX $obj->_root->{end} isn't updated from these 10 bytes that are being written
- my $plain_key = "[base]";
- print( $fh pack($self->{data_pack}, length($plain_key)) . $plain_key );
+ return unless $bytes_read;
- # Flush the filehandle
- my $old_fh = select $fh;
- my $old_af = $|; $| = 1; $| = $old_af;
- select $old_fh;
+ my ($file_signature, $sig_header, $header_version, $size) = unpack(
+ 'A4 A N N', $buffer
+ );
- return 1;
+ unless ( $file_signature eq SIG_FILE ) {
+ $self->_fileobj->close;
+ $self->_throw_error( "Signature not found -- file is not a Deep DB" );
}
- ##
- # Check signature was valid
- ##
- unless ($signature eq SIG_FILE) {
- $self->close_fh( $obj );
- $obj->_throw_error("Signature not found -- file is not a Deep DB");
+ unless ( $sig_header eq SIG_HEADER ) {
+ $self->_fileobj->close;
+ $self->_throw_error( "Old file version found." );
}
- ##
- # Get our type from master index signature
- ##
- my $tag = $self->load_tag($obj, $obj->_base_offset)
- or $obj->_throw_error("Corrupted file, no master index record");
+ my $buffer2;
+ $bytes_read += read( $fh, $buffer2, $size );
+ my ($running_transactions, @values) = unpack( 'N n A n A n', $buffer2 );
- unless ($obj->{type} eq $tag->{signature}) {
- $obj->_throw_error("File type mismatch");
+ $self->_fileobj->set_transaction_offset( 13 );
+
+ if ( @values < 5 || grep { !defined } @values ) {
+ $self->_fileobj->close;
+ $self->_throw_error("Corrupted file - bad header");
}
-#XXX We probably also want to store the hash algorithm name and not assume anything
-#XXX The cool thing would be to allow a different hashing algorithm at every level
+ #XXX Add warnings if values weren't set right
+ @{$self}{qw(long_size long_pack data_size data_pack max_buckets)} = @values;
- return 1;
+ return $bytes_read;
}
-sub close_fh {
+sub setup_fh {
my $self = shift;
my ($obj) = @_;
- if ( my $fh = $obj->_root->{fh} ) {
- close $fh;
+ local($/,$\);
+
+ my $fh = $self->_fh;
+ flock $fh, LOCK_EX;
+
+ #XXX The duplication of calculate_sizes needs to go away
+ unless ( $obj->{base_offset} ) {
+ my $bytes_read = $self->read_file_header;
+
+ $self->calculate_sizes;
+
+ ##
+ # File is empty -- write header and master index
+ ##
+ if (!$bytes_read) {
+ $self->_fileobj->audit( "# Database created on" );
+
+ $self->write_file_header;
+
+ $obj->{base_offset} = $self->_request_space( $self->tag_size( $self->{index_size} ) );
+
+ $self->write_tag(
+ $obj->_base_offset, $obj->_type,
+ chr(0)x$self->{index_size},
+ );
+
+ # Flush the filehandle
+ my $old_fh = select $fh;
+ my $old_af = $|; $| = 1; $| = $old_af;
+ select $old_fh;
+ }
+ else {
+ $obj->{base_offset} = $bytes_read;
+
+ ##
+ # Get our type from master index header
+ ##
+ my $tag = $self->load_tag($obj->_base_offset);
+ unless ( $tag ) {
+ flock $fh, LOCK_UN;
+ $self->_throw_error("Corrupted file, no master index record");
+ }
+
+ unless ($obj->_type eq $tag->{signature}) {
+ flock $fh, LOCK_UN;
+ $self->_throw_error("File type mismatch");
+ }
+ }
+ }
+ else {
+ $self->calculate_sizes;
+ }
+
+ #XXX We have to make sure we don't mess up when autoflush isn't turned on
+ unless ( $self->_fileobj->{inode} ) {
+ my @stats = stat($fh);
+ $self->_fileobj->{inode} = $stats[1];
+ $self->_fileobj->{end} = $stats[7];
}
- $obj->_root->{fh} = undef;
+
+ flock $fh, LOCK_UN;
return 1;
}
-sub create_tag {
+sub tag_size {
+ my $self = shift;
+ my ($size) = @_;
+ return SIG_SIZE + $self->{data_size} + $size;
+}
+
+sub write_tag {
##
# Given offset, signature and content, create tag and write to disk
##
my $self = shift;
- my ($obj, $offset, $sig, $content) = @_;
- my $size = length($content);
+ my ($offset, $sig, $content) = @_;
+ my $size = length( $content );
- my $fh = $obj->_fh;
+ local($/,$\);
- seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
- print( $fh $sig . pack($self->{data_pack}, $size) . $content );
+ my $fh = $self->_fh;
- if ($offset == $obj->_root->{end}) {
- $obj->_root->{end} += SIG_SIZE + $self->{data_size} + $size;
+ if ( defined $offset ) {
+ seek($fh, $offset + $self->_fileobj->{file_offset}, SEEK_SET);
}
+ print( $fh $sig . pack($self->{data_pack}, $size) . $content );
+
+ return unless defined $offset;
+
return {
signature => $sig,
size => $size,
# Given offset, load single tag and return signature, size and data
##
my $self = shift;
- my ($obj, $offset) = @_;
+ my ($offset) = @_;
+
+ local($/,$\);
- my $fh = $obj->_fh;
+# print join(':',map{$_||''}caller(1)), $/;
- seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET);
+ my $fh = $self->_fh;
+
+ seek($fh, $offset + $self->_fileobj->{file_offset}, SEEK_SET);
#XXX I'm not sure this check will work if autoflush isn't enabled ...
return if eof $fh;
};
}
+sub _get_dbm_object {
+ my $item = shift;
+
+ my $obj = eval {
+ local $SIG{__DIE__};
+ if ($item->isa( 'DBM::Deep' )) {
+ return $item;
+ }
+ return;
+ };
+ return $obj if $obj;
+
+ my $r = Scalar::Util::reftype( $item ) || '';
+ if ( $r eq 'HASH' ) {
+ my $obj = eval {
+ local $SIG{__DIE__};
+ my $obj = tied(%$item);
+ if ($obj->isa( 'DBM::Deep' )) {
+ return $obj;
+ }
+ return;
+ };
+ return $obj if $obj;
+ }
+ elsif ( $r eq 'ARRAY' ) {
+ my $obj = eval {
+ local $SIG{__DIE__};
+ my $obj = tied(@$item);
+ if ($obj->isa( 'DBM::Deep' )) {
+ return $obj;
+ }
+ return;
+ };
+ return $obj if $obj;
+ }
+
+ return;
+}
+
+sub _length_needed {
+ my $self = shift;
+ my ($value, $key) = @_;
+
+ my $is_dbm_deep = eval {
+ local $SIG{'__DIE__'};
+ $value->isa( 'DBM::Deep' );
+ };
+
+ my $len = SIG_SIZE + $self->{data_size}
+ + $self->{data_size} + length( $key );
+
+ if ( $is_dbm_deep && $value->_fileobj eq $self->_fileobj ) {
+ return $len + $self->{long_size};
+ }
+
+ my $r = Scalar::Util::reftype( $value ) || '';
+ if ( $self->_fileobj->{autobless} ) {
+ # This is for the bit saying whether or not this thing is blessed.
+ $len += 1;
+ }
+
+ unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
+ if ( defined $value ) {
+ $len += length( $value );
+ }
+ return $len;
+ }
+
+ $len += $self->{index_size};
+
+ # if autobless is enabled, must also take into consideration
+ # the class name as it is stored after the key.
+ if ( $self->_fileobj->{autobless} ) {
+ my $c = Scalar::Util::blessed($value);
+ if ( defined $c && !$is_dbm_deep ) {
+ $len += $self->{data_size} + length($c);
+ }
+ }
+
+ return $len;
+}
+
sub add_bucket {
##
# Adds one key/value pair to bucket list, given offset, MD5 digest of key,
# plain (undigested) key and value.
##
my $self = shift;
- my ($obj, $tag, $md5, $plain_key, $value) = @_;
-
- my $keys = $tag->{content};
- my $location = 0;
- my $result = 2;
+ my ($tag, $md5, $plain_key, $value, $deleted, $orig_key) = @_;
+ $deleted ||= 0;
- my $root = $obj->_root;
+ local($/,$\);
- my $is_dbm_deep = eval { local $SIG{'__DIE__'}; $value->isa( 'DBM::Deep' ) };
- my $internal_ref = $is_dbm_deep && ($value->_root eq $root);
+ # This verifies that only supported values will be stored.
+ {
+ my $r = Scalar::Util::reftype( $value );
+ last if !defined $r;
- my $fh = $obj->_fh;
+ last if $r eq 'HASH';
+ last if $r eq 'ARRAY';
- ##
- # Iterate through buckets, seeing if this is a new entry or a replace.
- ##
- BUCKET:
- for (my $i = 0; $i < $self->{max_buckets}; $i++) {
- my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
+ $self->_throw_error(
+ "Storage of variables of type '$r' is not supported."
+ );
+ }
- if (!$subloc) {
- ##
- # Found empty bucket (end of list). Populate and exit loop.
- ##
- $result = 2;
+ my $location = 0;
+ my $result = 2;
- $location = $root->{end};
+ my $root = $self->_fileobj;
+ my $fh = $self->_fh;
- seek(
- $fh,
- $tag->{offset} + ($i * $self->{bucket_size}) + $root->{file_offset},
- SEEK_SET,
- );
+ my $actual_length = $self->_length_needed( $value, $plain_key );
- print( $fh $md5 . pack($self->{long_pack}, $location) );
- last;
- }
+ #ACID - This is a mutation. Must only find the exact transaction
+ my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5, 1 );
- if ( $md5 ne $key ) {
- next BUCKET;
- }
+ my @transactions;
+ if ( $self->_fileobj->transaction_id == 0 ) {
+ @transactions = $self->_fileobj->current_transactions;
+ }
- ##
- # Found existing bucket with same key. Replace with new value.
- ##
+# $self->_release_space( $size, $subloc );
+ # Updating a known md5
+#XXX This needs updating to use _release_space
+ if ( $subloc ) {
$result = 1;
- seek($fh, $subloc + SIG_SIZE + $root->{file_offset}, SEEK_SET);
- my $size;
- read( $fh, $size, $self->{data_size});
- $size = unpack($self->{data_pack}, $size);
-
- ##
- # If value is a hash, array, or raw value with equal or less size, we can
- # reuse the same content area of the database. Otherwise, we have to create
- # a new content area at the EOF.
- ##
- my $actual_length;
- if ( $internal_ref ) {
- $actual_length = $self->{long_size};
- }
- else {
- my $r = Scalar::Util::reftype( $value ) || '';
- if ( $r eq 'HASH' || $r eq 'ARRAY' ) {
- $actual_length = $self->{index_size};
-
- # if autobless is enabled, must also take into consideration
- # the class name, as it is stored along with key/value.
- if ( $root->{autobless} ) {
- my $value_class = Scalar::Util::blessed($value);
- if ( defined $value_class && !$value->isa('DBM::Deep') ) {
- $actual_length += length($value_class);
- }
- }
- }
- else { $actual_length = length($value); }
- }
-
if ($actual_length <= $size) {
$location = $subloc;
}
else {
- $location = $root->{end};
+ $location = $self->_request_space( $actual_length );
seek(
$fh,
- $tag->{offset} + ($i * $self->{bucket_size}) + $self->{hash_size} + $root->{file_offset},
+ $tag->{offset} + $offset
+ + $self->{hash_size} + $root->{file_offset},
SEEK_SET,
);
- print( $fh pack($self->{long_pack}, $location) );
+ print( $fh pack($self->{long_pack}, $location ) );
+ print( $fh pack($self->{long_pack}, $actual_length ) );
+ print( $fh pack('n n', $root->transaction_id, $deleted ) );
+ }
+ }
+ # Adding a new md5
+ elsif ( defined $offset ) {
+ $location = $self->_request_space( $actual_length );
+
+ seek( $fh, $tag->{offset} + $offset + $root->{file_offset}, SEEK_SET );
+ print( $fh $md5 . pack($self->{long_pack}, $location ) );
+ print( $fh pack($self->{long_pack}, $actual_length ) );
+ print( $fh pack('n n', $root->transaction_id, $deleted ) );
+
+ for ( @transactions ) {
+ my $tag2 = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
+ $self->_fileobj->{transaction_id} = $_;
+ $self->add_bucket( $tag2, $md5, '', '', 1, $orig_key );
+ $self->_fileobj->{transaction_id} = 0;
}
-
- last;
}
-
- ##
# If bucket didn't fit into list, split into a new index level
- ##
- if (!$location) {
- $self->split_index( $obj, $md5, $tag );
+ # split_index() will do the _request_space() call
+ else {
+ $location = $self->split_index( $md5, $tag );
+ }
+
+ $self->write_value( $location, $plain_key, $value, $orig_key );
+
+ return $result;
+}
+
+sub write_value {
+ my $self = shift;
+ my ($location, $key, $value, $orig_key) = @_;
- $location = $root->{end};
+ local($/,$\);
+
+ my $fh = $self->_fh;
+ my $root = $self->_fileobj;
+
+ my $dbm_deep_obj = _get_dbm_object( $value );
+ if ( $dbm_deep_obj && $dbm_deep_obj->_fileobj ne $self->_fileobj ) {
+ $self->_throw_error( "Cannot cross-reference. Use export() instead" );
}
+ seek($fh, $location + $root->{file_offset}, SEEK_SET);
+
##
- # Seek to content area and store signature, value and plaintext key
+ # Write signature based on content type, set content length and write
+ # actual value.
##
- if ($location) {
- seek($fh, $location + $root->{file_offset}, SEEK_SET);
-
- ##
- # Write signature based on content type, set content length and write
- # actual value.
- ##
- my $r = Scalar::Util::reftype($value) || '';
- my $content_length;
- if ( $internal_ref ) {
- print( $fh SIG_INTERNAL );
- print( $fh pack($self->{data_pack}, $self->{long_size}) );
- print( $fh pack($self->{long_pack}, $value->_base_offset) );
- $content_length = $self->{long_size};
+ my $r = Scalar::Util::reftype( $value ) || '';
+ if ( $dbm_deep_obj ) {
+ $self->write_tag( undef, SIG_INTERNAL,pack($self->{long_pack}, $dbm_deep_obj->_base_offset) );
+ }
+ elsif ($r eq 'HASH') {
+ if ( !$dbm_deep_obj && tied %{$value} ) {
+ $self->_throw_error( "Cannot store something that is tied" );
}
- else {
- if ($r eq 'HASH') {
- print( $fh SIG_HASH );
- print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} );
- $content_length = $self->{index_size};
- }
- elsif ($r eq 'ARRAY') {
- print( $fh SIG_ARRAY );
- print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} );
- $content_length = $self->{index_size};
- }
- elsif (!defined($value)) {
- print( $fh SIG_NULL );
- print( $fh pack($self->{data_pack}, 0) );
- $content_length = 0;
- }
- else {
- print( $fh SIG_DATA );
- print( $fh pack($self->{data_pack}, length($value)) . $value );
- $content_length = length($value);
- }
+ $self->write_tag( undef, SIG_HASH, chr(0)x$self->{index_size} );
+ }
+ elsif ($r eq 'ARRAY') {
+ if ( !$dbm_deep_obj && tied @{$value} ) {
+ $self->_throw_error( "Cannot store something that is tied" );
}
+ $self->write_tag( undef, SIG_ARRAY, chr(0)x$self->{index_size} );
+ }
+ elsif (!defined($value)) {
+ $self->write_tag( undef, SIG_NULL, '' );
+ }
+ else {
+ $self->write_tag( undef, SIG_DATA, $value );
+ }
- ##
- # Plain key is stored AFTER value, as keys are typically fetched less often.
- ##
- print( $fh pack($self->{data_pack}, length($plain_key)) . $plain_key );
+ ##
+ # Plain key is stored AFTER value, as keys are typically fetched less often.
+ ##
+ print( $fh pack($self->{data_pack}, length($key)) . $key );
- ##
- # If value is blessed, preserve class name
- ##
- if ( $root->{autobless} ) {
- my $value_class = Scalar::Util::blessed($value);
- if ( defined $value_class && !$value->isa( 'DBM::Deep' ) ) {
- ##
- # Blessed ref -- will restore later
- ##
- print( $fh chr(1) );
- print( $fh pack($self->{data_pack}, length($value_class)) . $value_class );
- $content_length += 1;
- $content_length += $self->{data_size} + length($value_class);
- }
- else {
- print( $fh chr(0) );
- $content_length += 1;
- }
- }
+ # Internal references don't care about autobless
+ return 1 if $dbm_deep_obj;
- ##
- # If this is a new content area, advance EOF counter
- ##
- if ($location == $root->{end}) {
- $root->{end} += SIG_SIZE;
- $root->{end} += $self->{data_size} + $content_length;
- $root->{end} += $self->{data_size} + length($plain_key);
+ ##
+ # If value is blessed, preserve class name
+ ##
+ if ( $root->{autobless} ) {
+ my $c = Scalar::Util::blessed($value);
+ if ( defined $c && !$dbm_deep_obj ) {
+ print( $fh chr(1) );
+ print( $fh pack($self->{data_pack}, length($c)) . $c );
}
-
- ##
- # If content is a hash or array, create new child DBM::Deep object and
- # pass each key or element to it.
- ##
- if ( ! $internal_ref ) {
- if ($r eq 'HASH') {
- my $branch = DBM::Deep->new(
- type => DBM::Deep->TYPE_HASH,
- base_offset => $location,
- root => $root,
- );
- foreach my $key (keys %{$value}) {
- $branch->STORE( $key, $value->{$key} );
- }
- }
- elsif ($r eq 'ARRAY') {
- my $branch = DBM::Deep->new(
- type => DBM::Deep->TYPE_ARRAY,
- base_offset => $location,
- root => $root,
- );
- my $index = 0;
- foreach my $element (@{$value}) {
- $branch->STORE( $index, $element );
- $index++;
- }
- }
+ else {
+ print( $fh chr(0) );
}
+ }
- return $result;
+ ##
+ # Tie the passed in reference so that changes to it are reflected in the
+ # datafile. The use of $location as the base_offset will act as the
+ # the linkage between parent and child.
+ #
+ # The overall assignment is a hack around the fact that just tying doesn't
+ # store the values. This may not be the wrong thing to do.
+ ##
+ if ($r eq 'HASH') {
+ my %x = %$value;
+ tie %$value, 'DBM::Deep', {
+ base_offset => $location,
+ fileobj => $root,
+ parent => $self->{obj},
+ parent_key => $orig_key,
+ };
+ %$value = %x;
+ }
+ elsif ($r eq 'ARRAY') {
+ my @x = @$value;
+ tie @$value, 'DBM::Deep', {
+ base_offset => $location,
+ fileobj => $root,
+ parent => $self->{obj},
+ parent_key => $orig_key,
+ };
+ @$value = @x;
}
- $obj->_throw_error("Fatal error: indexing failed -- possibly due to corruption in file");
+ return 1;
}
sub split_index {
my $self = shift;
- my ($obj, $md5, $tag) = @_;
+ my ($md5, $tag) = @_;
- my $fh = $obj->_fh;
- my $root = $obj->_root;
- my $keys = $tag->{content};
+ local($/,$\);
+
+ my $fh = $self->_fh;
+ my $root = $self->_fileobj;
+
+ my $loc = $self->_request_space(
+ $self->tag_size( $self->{index_size} ),
+ );
seek($fh, $tag->{ref_loc} + $root->{file_offset}, SEEK_SET);
- print( $fh pack($self->{long_pack}, $root->{end}) );
+ print( $fh pack($self->{long_pack}, $loc) );
- my $index_tag = $self->create_tag(
- $obj,
- $root->{end},
- SIG_INDEX,
- chr(0) x $self->{index_size},
+ my $index_tag = $self->write_tag(
+ $loc, SIG_INDEX,
+ chr(0)x$self->{index_size},
);
- my @offsets = ();
+ my $newtag_loc = $self->_request_space(
+ $self->tag_size( $self->{bucket_list_size} ),
+ );
- $keys .= $md5 . pack($self->{long_pack}, 0);
+ my $keys = $tag->{content}
+ . $md5 . pack($self->{long_pack}, $newtag_loc)
+ . pack($self->{long_pack}, 0) # size
+ . pack($self->{long_pack}, 0); # transaction ID
+ my @newloc = ();
BUCKET:
for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
- my ($key, $old_subloc) = $self->_get_key_subloc( $keys, $i );
+ my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
- next BUCKET unless $key;
+ die "[INTERNAL ERROR]: No key in split_index()\n" unless $key;
+ die "[INTERNAL ERROR]: No subloc in split_index()\n" unless $old_subloc;
my $num = ord(substr($key, $tag->{ch} + 1, 1));
- if ($offsets[$num]) {
- my $offset = $offsets[$num] + SIG_SIZE + $self->{data_size};
- seek($fh, $offset + $root->{file_offset}, SEEK_SET);
+ if ($newloc[$num]) {
+ seek($fh, $newloc[$num] + $root->{file_offset}, SEEK_SET);
my $subkeys;
read( $fh, $subkeys, $self->{bucket_list_size});
- for (my $k=0; $k<$self->{max_buckets}; $k++) {
- my ($temp, $subloc) = $self->_get_key_subloc( $subkeys, $k );
+ # This is looking for the first empty spot
+ my ($subloc, $offset, $size) = $self->_find_in_buckets(
+ { content => $subkeys }, '',
+ );
+
+ seek($fh, $newloc[$num] + $offset + $root->{file_offset}, SEEK_SET);
+ print( $fh $key . pack($self->{long_pack}, $old_subloc) );
- if (!$subloc) {
- seek($fh, $offset + ($k * $self->{bucket_size}) + $root->{file_offset}, SEEK_SET);
- print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
- last;
- }
- } # k loop
+ next;
}
- else {
- $offsets[$num] = $root->{end};
- seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET);
- print( $fh pack($self->{long_pack}, $root->{end}) );
- my $blist_tag = $self->create_tag($obj, $root->{end}, SIG_BLIST, chr(0) x $self->{bucket_list_size});
+ seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET);
- seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET);
- print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) );
- }
- } # i loop
+ my $loc = $self->_request_space(
+ $self->tag_size( $self->{bucket_list_size} ),
+ );
- return;
+ print( $fh pack($self->{long_pack}, $loc) );
+
+ my $blist_tag = $self->write_tag(
+ $loc, SIG_BLIST,
+ chr(0)x$self->{bucket_list_size},
+ );
+
+ seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET);
+ print( $fh $key . pack($self->{long_pack}, $old_subloc) );
+
+ $newloc[$num] = $blist_tag->{offset};
+ }
+
+ $self->_release_space(
+ $self->tag_size( $self->{bucket_list_size} ),
+ $tag->{offset} - SIG_SIZE - $self->{data_size},
+ );
+
+ return $newtag_loc;
}
sub read_from_loc {
my $self = shift;
- my ($obj, $subloc) = @_;
+ my ($subloc, $orig_key) = @_;
- my $fh = $obj->_fh;
+ local($/,$\);
+
+ my $fh = $self->_fh;
##
# Found match -- seek to offset and read signature
##
my $signature;
- seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
+ seek($fh, $subloc + $self->_fileobj->{file_offset}, SEEK_SET);
read( $fh, $signature, SIG_SIZE);
##
# If value is a hash or array, return new DBM::Deep object with correct offset
##
if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
- my $obj = DBM::Deep->new(
- type => $signature,
+ my $new_obj = DBM::Deep->new({
+ type => $signature,
base_offset => $subloc,
- root => $obj->_root,
- );
+ fileobj => $self->_fileobj,
+ parent => $self->{obj},
+ parent_key => $orig_key,
+ });
- if ($obj->_root->{autobless}) {
+ if ($new_obj->_fileobj->{autobless}) {
##
# Skip over value and plain key to see if object needs
# to be re-blessed
seek($fh, $self->{data_size} + $self->{index_size}, SEEK_CUR);
my $size;
- read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
+ read( $fh, $size, $self->{data_size});
+ $size = unpack($self->{data_pack}, $size);
if ($size) { seek($fh, $size, SEEK_CUR); }
my $bless_bit;
# Yes, object needs to be re-blessed
##
my $class_name;
- read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size);
+ read( $fh, $size, $self->{data_size});
+ $size = unpack($self->{data_pack}, $size);
if ($size) { read( $fh, $class_name, $size); }
- if ($class_name) { $obj = bless( $obj, $class_name ); }
+ if ($class_name) { $new_obj = bless( $new_obj, $class_name ); }
}
}
- return $obj;
+ return $new_obj;
}
elsif ( $signature eq SIG_INTERNAL ) {
my $size;
read( $fh, $new_loc, $size );
$new_loc = unpack( $self->{long_pack}, $new_loc );
- return $self->read_from_loc( $obj, $new_loc );
+ return $self->read_from_loc( $new_loc, $orig_key );
}
else {
return;
##
# Otherwise return actual value
##
- elsif ($signature eq SIG_DATA) {
+ elsif ( $signature eq SIG_DATA ) {
my $size;
read( $fh, $size, $self->{data_size});
$size = unpack($self->{data_pack}, $size);
# Fetch single value given tag and MD5 digested key.
##
my $self = shift;
- my ($obj, $tag, $md5) = @_;
- my $keys = $tag->{content};
-
- my $fh = $obj->_fh;
-
- ##
- # Iterate through buckets, looking for a key match
- ##
- BUCKET:
- for (my $i = 0; $i < $self->{max_buckets}; $i++) {
- my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
-
- if (!$subloc) {
- ##
- # Hit end of list, no match
- ##
- return;
- }
-
- if ( $md5 ne $key ) {
- next BUCKET;
- }
-
- return $self->read_from_loc( $obj, $subloc );
- } # i loop
+ my ($tag, $md5, $orig_key) = @_;
+ #ACID - This is a read. Can find exact or HEAD
+ my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5 );
+ if ( $subloc && !$is_deleted ) {
+ return $self->read_from_loc( $subloc, $orig_key );
+ }
return;
}
# Delete single key/value pair given tag and MD5 digested key.
##
my $self = shift;
- my ($obj, $tag, $md5) = @_;
- my $keys = $tag->{content};
-
- my $fh = $obj->_fh;
-
- ##
- # Iterate through buckets, looking for a key match
- ##
- BUCKET:
- for (my $i=0; $i<$self->{max_buckets}; $i++) {
- my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
-
- if (!$subloc) {
- ##
- # Hit end of list, no match
- ##
- return;
- }
+ my ($tag, $md5, $orig_key) = @_;
- if ( $md5 ne $key ) {
- next BUCKET;
- }
+ local($/,$\);
- ##
- # Matched key -- delete bucket and return
- ##
- seek($fh, $tag->{offset} + ($i * $self->{bucket_size}) + $obj->_root->{file_offset}, SEEK_SET);
- print( $fh substr($keys, ($i+1) * $self->{bucket_size} ) );
+ #ACID - This is a mutation. Must only find the exact transaction
+ my ($subloc, $offset, $size) = $self->_find_in_buckets( $tag, $md5, 1 );
+#XXX This needs _release_space()
+ if ( $subloc ) {
+ my $fh = $self->_fh;
+ seek($fh, $tag->{offset} + $offset + $self->_fileobj->{file_offset}, SEEK_SET);
+ print( $fh substr($tag->{content}, $offset + $self->{bucket_size} ) );
print( $fh chr(0) x $self->{bucket_size} );
return 1;
- } # i loop
-
+ }
return;
}
# Check existence of single key given tag and MD5 digested key.
##
my $self = shift;
- my ($obj, $tag, $md5) = @_;
- my $keys = $tag->{content};
+ my ($tag, $md5) = @_;
- ##
- # Iterate through buckets, looking for a key match
- ##
- BUCKET:
- for (my $i=0; $i<$self->{max_buckets}; $i++) {
- my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
-
- if (!$subloc) {
- ##
- # Hit end of list, no match
- ##
- return;
- }
-
- if ( $md5 ne $key ) {
- next BUCKET;
- }
-
- ##
- # Matched key -- return true
- ##
- return 1;
- } # i loop
-
- return;
+ #ACID - This is a read. Can find exact or HEAD
+ my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
+ return ($subloc && !$is_deleted) && 1;
}
sub find_bucket_list {
# Locate offset for bucket list, given digested key
##
my $self = shift;
- my ($obj, $md5, $args) = @_;
+ my ($offset, $md5, $args) = @_;
$args = {} unless $args;
+ local($/,$\);
+
##
# Locate offset for bucket list using digest index system
##
- my $tag = $self->load_tag($obj, $obj->_base_offset)
+ my $tag = $self->load_tag( $offset )
or $self->_throw_error( "INTERNAL ERROR - Cannot find tag" );
my $ch = 0;
my $num = ord substr($md5, $ch, 1);
my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
- $tag = $self->index_lookup( $obj, $tag, $num );
+ $tag = $self->index_lookup( $tag, $num );
if (!$tag) {
- if ( $args->{create} ) {
- my $fh = $obj->_fh;
- seek($fh, $ref_loc + $obj->_root->{file_offset}, SEEK_SET);
- print( $fh pack($self->{long_pack}, $obj->_root->{end}) );
-
- $tag = $self->create_tag(
- $obj, $obj->_root->{end},
- SIG_BLIST,
- chr(0) x $self->{bucket_list_size},
- );
+ return if !$args->{create};
- $tag->{ref_loc} = $ref_loc;
- $tag->{ch} = $ch;
+ my $loc = $self->_request_space(
+ $self->tag_size( $self->{bucket_list_size} ),
+ );
- last;
- }
- else {
- return;
- }
+ my $fh = $self->_fh;
+ seek($fh, $ref_loc + $self->_fileobj->{file_offset}, SEEK_SET);
+ print( $fh pack($self->{long_pack}, $loc) );
+
+ $tag = $self->write_tag(
+ $loc, SIG_BLIST,
+ chr(0)x$self->{bucket_list_size},
+ );
+
+ $tag->{ref_loc} = $ref_loc;
+ $tag->{ch} = $ch;
+
+ last;
}
- $tag->{ch} = $ch;
+ $tag->{ch} = $ch++;
$tag->{ref_loc} = $ref_loc;
-
- $ch++;
}
return $tag;
# Given index tag, lookup single entry in index and return .
##
my $self = shift;
- my ($obj, $tag, $index) = @_;
+ my ($tag, $index) = @_;
my $location = unpack(
$self->{long_pack},
if (!$location) { return; }
- return $self->load_tag( $obj, $location );
+ return $self->load_tag( $location );
}
sub traverse_index {
my $self = shift;
my ($obj, $offset, $ch, $force_return_next) = @_;
- my $tag = $self->load_tag($obj, $offset );
+ local($/,$\);
+
+ my $tag = $self->load_tag( $offset );
- my $fh = $obj->_fh;
+ my $fh = $self->_fh;
if ($tag->{signature} ne SIG_BLIST) {
my $content = $tag->{content};
my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
- for (my $index = $start; $index < 256; $index++) {
+ for (my $idx = $start; $idx < (2**8); $idx++) {
my $subloc = unpack(
$self->{long_pack},
- substr($content, $index * $self->{long_size}, $self->{long_size}),
+ substr(
+ $content,
+ $idx * $self->{long_size},
+ $self->{long_size},
+ ),
);
if ($subloc) {
}
# Seek to bucket location and skip over signature
elsif ($obj->{return_next}) {
- seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET);
+ seek($fh, $subloc + $self->_fileobj->{file_offset}, SEEK_SET);
# Skip over value to get to plain key
my $sig;
my $self = shift;
my ($keys, $idx) = @_;
- my ($key, $subloc) = unpack(
- "a$self->{hash_size} $self->{long_pack}",
+ my ($key, $subloc, $size, $transaction_id, $is_deleted) = unpack(
+ # This is 'a', not 'A'. Please read the pack() documentation for the
+ # difference between the two and why it's important.
+ "a$self->{hash_size} $self->{long_pack}2 n2",
substr(
$keys,
($idx * $self->{bucket_size}),
),
);
- return ($key, $subloc);
+ return ($key, $subloc, $size, $transaction_id, $is_deleted);
+}
+
+sub _find_in_buckets {
+ my $self = shift;
+ my ($tag, $md5, $exact) = @_;
+
+ my $trans_id = $self->_fileobj->transaction_id;
+
+ my @zero;
+
+ BUCKET:
+ for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
+ my ($key, $subloc, $size, $transaction_id, $is_deleted) = $self->_get_key_subloc(
+ $tag->{content}, $i,
+ );
+
+ my @rv = ($subloc, $i * $self->{bucket_size}, $size, $is_deleted);
+
+ unless ( $subloc ) {
+ if ( !$exact && @zero and $trans_id ) {
+ @rv = ($zero[2], $zero[0] * $self->{bucket_size},$zero[3],$is_deleted);
+ }
+ return @rv;
+ }
+
+ next BUCKET if $key ne $md5;
+
+ # Save off the HEAD in case we need it.
+ @zero = ($i,$key,$subloc,$size,$transaction_id,$is_deleted) if $transaction_id == 0;
+
+ next BUCKET if $transaction_id != $trans_id;
+
+ return @rv;
+ }
+
+ return;
+}
+
+sub _request_space {
+ my $self = shift;
+ my ($size) = @_;
+
+ my $loc = $self->_fileobj->{end};
+ $self->_fileobj->{end} += $size;
+
+ return $loc;
+}
+
+sub _release_space {
+ my $self = shift;
+ my ($size, $loc) = @_;
+
+ local($/,$\);
+
+ my $next_loc = 0;
+
+ my $fh = $self->_fh;
+ seek( $fh, $loc + $self->_fileobj->{file_offset}, SEEK_SET );
+ print( $fh SIG_FREE
+ . pack($self->{long_pack}, $size )
+ . pack($self->{long_pack}, $next_loc )
+ );
+
+ return;
+}
+
+sub _throw_error {
+ die "DBM::Deep: $_[1]\n";
}
1;
__END__
+
+# This will be added in later, after more refactoring is done. This is an early
+# attempt at refactoring on the physical level instead of the virtual level.
+sub _read_at {
+ my $self = shift;
+ my ($spot, $amount, $unpack) = @_;
+
+ local($/,$\);
+
+ my $fh = $self->_fh;
+ seek( $fh, $spot + $self->_fileobj->{file_offset}, SEEK_SET );
+
+ my $buffer;
+ my $bytes_read = read( $fh, $buffer, $amount );
+
+ if ( $unpack ) {
+ $buffer = unpack( $unpack, $buffer );
+ }
+
+ if ( wantarray ) {
+ return ($buffer, $bytes_read);
+ }
+ else {
+ return $buffer;
+ }
+}
+
+sub _print_at {
+ my $self = shift;
+ my ($spot, $data) = @_;
+
+ local($/,$\);
+
+ my $fh = $self->_fh;
+ seek( $fh, $spot, SEEK_SET );
+ print( $fh $data );
+
+ return;
+}
+
+sub get_file_version {
+ my $self = shift;
+
+ local($/,$\);
+
+ my $fh = $self->_fh;
+
+ seek( $fh, 13 + $self->_fileobj->{file_offset}, SEEK_SET );
+ my $buffer;
+ my $bytes_read = read( $fh, $buffer, 4 );
+ unless ( $bytes_read == 4 ) {
+ $self->_throw_error( "Cannot read file version" );
+ }
+
+ return unpack( 'N', $buffer );
+}
+
+sub write_file_version {
+ my $self = shift;
+ my ($new_version) = @_;
+
+ local($/,$\);
+
+ my $fh = $self->_fh;
+
+ seek( $fh, 13 + $self->_fileobj->{file_offset}, SEEK_SET );
+ print( $fh pack( 'N', $new_version ) );
+
+ return;
+}
+