package DBM::Deep::Engine;
-use 5.6.0;
+use 5.006_000;
use strict;
use warnings;
-use Fcntl qw( :DEFAULT :flock :seek );
+our $VERSION = q(1.0000);
+
use Scalar::Util ();
# File-wide notes:
-# * To add to bucket_size, make sure you modify the following:
-# - calculate_sizes()
-# - _get_key_subloc()
-# - add_bucket() - where the buckets are printed
+# * Every method in here assumes that the storage has been appropriately
+# safeguarded. This can be anything from flock() to some sort of manual
+# mutex. But, it's the caller's responsability to make sure that this has
+# been done.
-##
# 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_NULL () { 'N' }
sub SIG_FREE () { 'F' }
sub SIG_SIZE () { 1 }
+my $STALE_SIZE = 2;
+
+# Please refer to the pack() documentation for further information
+my %StP = (
+ 1 => 'C', # Unsigned char value (no order needed as it's just one byte)
+ 2 => 'n', # Unsigned short in "network" (big-endian) order
+ 4 => 'N', # Unsigned long in "network" (big-endian) order
+ 8 => 'Q', # Usigned quad (no order specified, presumably machine-dependent)
+);
+
+################################################################################
+
sub new {
my $class = shift;
my ($args) = @_;
my $self = bless {
- long_size => 4,
- long_pack => 'N',
- data_size => 4,
- data_pack => 'N',
-
- digest => \&Digest::MD5::md5,
- 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.
- ##
+ byte_size => 4,
+
+ digest => undef,
+ hash_size => 16, # In bytes
+ hash_chars => 256, # Number of chars the algorithm uses per byte
max_buckets => 16,
+ num_txns => 1, # The HEAD
+ trans_id => 0, # Default to the HEAD
+
+ data_sector_size => 64, # Size in bytes of each data sector
- fileobj => undef,
- obj => undef,
+ entries => {}, # This is the list of entries for transactions
+ storage => undef,
}, $class;
+ # Never allow byte_size to be set directly.
+ delete $args->{byte_size};
if ( defined $args->{pack_size} ) {
if ( lc $args->{pack_size} eq 'small' ) {
- $args->{long_size} = 2;
- $args->{long_pack} = 'n';
+ $args->{byte_size} = 2;
}
elsif ( lc $args->{pack_size} eq 'medium' ) {
- $args->{long_size} = 4;
- $args->{long_pack} = 'N';
+ $args->{byte_size} = 4;
}
elsif ( lc $args->{pack_size} eq 'large' ) {
- $args->{long_size} = 8;
- $args->{long_pack} = 'Q';
+ $args->{byte_size} = 8;
}
else {
- die "Unknown pack_size value: '$args->{pack_size}'\n";
+ DBM::Deep->_throw_error( "Unknown pack_size value: '$args->{pack_size}'" );
}
}
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";
- $self->{max_buckets} = 16;
+ my %validations = (
+ max_buckets => { floor => 16, ceil => 256 },
+ num_txns => { floor => 1, ceil => 255 },
+ data_sector_size => { floor => 32, ceil => 256 },
+ );
+
+ while ( my ($attr, $c) = each %validations ) {
+ if ( !defined $self->{$attr}
+ || !length $self->{$attr}
+ || $self->{$attr} =~ /\D/
+ || $self->{$attr} < $c->{floor}
+ ) {
+ $self->{$attr} = '(undef)' if !defined $self->{$attr};
+ warn "Floor of $attr is $c->{floor}. Setting it to $c->{floor} from '$self->{$attr}'\n";
+ $self->{$attr} = $c->{floor};
+ }
+ elsif ( $self->{$attr} > $c->{ceil} ) {
+ warn "Ceiling of $attr is $c->{ceil}. Setting it to $c->{ceil} from '$self->{$attr}'\n";
+ $self->{$attr} = $c->{ceil};
+ }
+ }
+
+ if ( !$self->{digest} ) {
+ require Digest::MD5;
+ $self->{digest} = \&Digest::MD5::md5;
}
return $self;
}
-sub _fileobj { return $_[0]{fileobj} }
+################################################################################
-sub calculate_sizes {
+sub read_value {
my $self = shift;
+ my ($obj, $key) = @_;
- #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};
+ # This will be a Reference sector
+ my $sector = $self->_load_sector( $obj->_base_offset )
+ or return;
- return;
-}
+ if ( $sector->staleness != $obj->_staleness ) {
+ return;
+ }
-sub write_file_header {
- my $self = shift;
+ my $key_md5 = $self->_apply_digest( $key );
- my $loc = $self->_fileobj->request_space( length( SIG_FILE ) + 21 );
+ my $value_sector = $sector->get_data_for({
+ key_md5 => $key_md5,
+ allow_head => 1,
+ });
- $self->_fileobj->print_at( $loc,
- 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}),
- );
+ unless ( $value_sector ) {
+ $value_sector = DBM::Deep::Engine::Sector::Null->new({
+ engine => $self,
+ data => undef,
+ });
- $self->_fileobj->set_transaction_offset( 13 );
+ $sector->write_data({
+ key_md5 => $key_md5,
+ key => $key,
+ value => $value_sector,
+ });
+ }
- return;
+ return $value_sector->data;
}
-sub read_file_header {
+sub get_classname {
my $self = shift;
+ my ($obj) = @_;
- my $buffer = $self->_fileobj->read_at( 0, length(SIG_FILE) + 9 );
- return unless length($buffer);
-
- my ($file_signature, $sig_header, $header_version, $size) = unpack(
- 'A4 A N N', $buffer
- );
+ # This will be a Reference sector
+ my $sector = $self->_load_sector( $obj->_base_offset )
+ or DBM::Deep->_throw_error( "How did get_classname fail (no sector for '$obj')?!" );
- unless ( $file_signature eq SIG_FILE ) {
- $self->_fileobj->close;
- $self->_throw_error( "Signature not found -- file is not a Deep DB" );
+ if ( $sector->staleness != $obj->_staleness ) {
+ return;
}
- unless ( $sig_header eq SIG_HEADER ) {
- $self->_fileobj->close;
- $self->_throw_error( "Old file version found." );
- }
+ return $sector->get_classname;
+}
- my $buffer2 = $self->_fileobj->read_at( undef, $size );
- my ($running_transactions, @values) = unpack( 'N n A n A n', $buffer2 );
+sub key_exists {
+ my $self = shift;
+ my ($obj, $key) = @_;
- $self->_fileobj->set_transaction_offset( 13 );
+ # This will be a Reference sector
+ my $sector = $self->_load_sector( $obj->_base_offset )
+ or return '';
- if ( @values < 5 || grep { !defined } @values ) {
- $self->_fileobj->close;
- $self->_throw_error("Corrupted file - bad header");
+ if ( $sector->staleness != $obj->_staleness ) {
+ return '';
}
- #XXX Add warnings if values weren't set right
- @{$self}{qw(long_size long_pack data_size data_pack max_buckets)} = @values;
+ my $data = $sector->get_data_for({
+ key_md5 => $self->_apply_digest( $key ),
+ allow_head => 1,
+ });
- return length($buffer) + length($buffer2);
+ # exists() returns 1 or '' for true/false.
+ return $data ? 1 : '';
}
-sub setup_fh {
+sub delete_key {
my $self = shift;
- my ($obj) = @_;
-
- # Need to remove use of $fh here
- my $fh = $self->_fileobj->{fh};
- flock $fh, LOCK_EX;
+ my ($obj, $key) = @_;
- #XXX The duplication of calculate_sizes needs to go away
- unless ( $obj->{base_offset} ) {
- my $bytes_read = $self->read_file_header;
+ my $sector = $self->_load_sector( $obj->_base_offset )
+ or return;
- $self->calculate_sizes;
+ if ( $sector->staleness != $obj->_staleness ) {
+ return;
+ }
- ##
- # File is empty -- write header and master index
- ##
- if (!$bytes_read) {
- $self->_fileobj->audit( "# Database created on" );
+ return $sector->delete_key({
+ key_md5 => $self->_apply_digest( $key ),
+ allow_head => 0,
+ });
+}
- $self->write_file_header;
+sub write_value {
+ my $self = shift;
+ my ($obj, $key, $value) = @_;
- $obj->{base_offset} = $self->_fileobj->request_space( $self->tag_size( $self->{index_size} ) );
+ my $r = Scalar::Util::reftype( $value ) || '';
+ {
+ last if $r eq '';
+ last if $r eq 'HASH';
+ last if $r eq 'ARRAY';
- $self->write_tag(
- $obj->_base_offset, $obj->_type,
- chr(0)x$self->{index_size},
- );
+ DBM::Deep->_throw_error(
+ "Storage of references of type '$r' is not supported."
+ );
+ }
- # Flush the filehandle
- my $old_fh = select $fh;
- my $old_af = $|; $| = 1; $| = $old_af;
- select $old_fh;
+ my ($class, $type);
+ if ( !defined $value ) {
+ $class = 'DBM::Deep::Engine::Sector::Null';
+ }
+ elsif ( $r eq 'ARRAY' || $r eq 'HASH' ) {
+ if ( $r eq 'ARRAY' && tied(@$value) ) {
+ DBM::Deep->_throw_error( "Cannot store something that is tied." );
}
- 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");
- }
+ if ( $r eq 'HASH' && tied(%$value) ) {
+ DBM::Deep->_throw_error( "Cannot store something that is tied." );
}
+ $class = 'DBM::Deep::Engine::Sector::Reference';
+ $type = substr( $r, 0, 1 );
}
else {
- $self->calculate_sizes;
+ $class = 'DBM::Deep::Engine::Sector::Scalar';
}
- #XXX We have to make sure we don't mess up when autoflush isn't turned on
- $self->_fileobj->set_inode;
+ # This will be a Reference sector
+ my $sector = $self->_load_sector( $obj->_base_offset )
+ or DBM::Deep->_throw_error( "Cannot write to a deleted spot in DBM::Deep." );
- flock $fh, LOCK_UN;
+ if ( $sector->staleness != $obj->_staleness ) {
+ DBM::Deep->_throw_error( "Cannot write to a deleted spot in DBM::Deep.n" );
+ }
- return 1;
-}
+ # Create this after loading the reference sector in case something bad happens.
+ # This way, we won't allocate value sector(s) needlessly.
+ my $value_sector = $class->new({
+ engine => $self,
+ data => $value,
+ type => $type,
+ });
+
+ $sector->write_data({
+ key => $key,
+ key_md5 => $self->_apply_digest( $key ),
+ value => $value_sector,
+ });
+
+ # This code is to make sure we write all the values in the $value to the disk
+ # and to make sure all changes to $value after the assignment are reflected
+ # on disk. This may be counter-intuitive at first, but it is correct dwimmery.
+ # NOTE - simply tying $value won't perform a STORE on each value. Hence, the
+ # copy to a temp value.
+ if ( $r eq 'ARRAY' ) {
+ my @temp = @$value;
+ tie @$value, 'DBM::Deep', {
+ base_offset => $value_sector->offset,
+ staleness => $value_sector->staleness,
+ storage => $self->storage,
+ engine => $self,
+ };
+ @$value = @temp;
+ bless $value, 'DBM::Deep::Array' unless Scalar::Util::blessed( $value );
+ }
+ elsif ( $r eq 'HASH' ) {
+ my %temp = %$value;
+ tie %$value, 'DBM::Deep', {
+ base_offset => $value_sector->offset,
+ staleness => $value_sector->staleness,
+ storage => $self->storage,
+ engine => $self,
+ };
-sub tag_size {
- my $self = shift;
- my ($size) = @_;
- return SIG_SIZE + $self->{data_size} + $size;
+ %$value = %temp;
+ bless $value, 'DBM::Deep::Hash' unless Scalar::Util::blessed( $value );
+ }
+
+ return 1;
}
-sub write_tag {
- ##
- # Given offset, signature and content, create tag and write to disk
- ##
+# XXX Add staleness here
+sub get_next_key {
my $self = shift;
- my ($offset, $sig, $content) = @_;
- my $size = length( $content );
+ my ($obj, $prev_key) = @_;
- $self->_fileobj->print_at(
- $offset,
- $sig, pack($self->{data_pack}, $size), $content,
- );
-
- return unless defined $offset;
+ # XXX Need to add logic about resetting the iterator if any key in the reference has changed
+ unless ( $prev_key ) {
+ $obj->{iterator} = DBM::Deep::Iterator->new({
+ base_offset => $obj->_base_offset,
+ engine => $self,
+ });
+ }
- return {
- signature => $sig,
- size => $size,
- offset => $offset + SIG_SIZE + $self->{data_size},
- content => $content
- };
+ return $obj->{iterator}->get_next_key( $obj );
}
-sub load_tag {
- ##
- # Given offset, load single tag and return signature, size and data
- ##
- my $self = shift;
- my ($offset) = @_;
+################################################################################
- my $fileobj = $self->_fileobj;
-
- my $s = SIG_SIZE + $self->{data_size};
- my $b = $fileobj->read_at( $offset, $s );
- my ($sig, $size) = unpack( "A $self->{data_pack}", $b );
+sub setup_fh {
+ my $self = shift;
+ my ($obj) = @_;
- my $buffer = $fileobj->read_at( undef, $size );
+ # We're opening the file.
+ unless ( $obj->_base_offset ) {
+ my $bytes_read = $self->_read_file_header;
- return {
- signature => $sig,
- size => $size,
- offset => $offset + SIG_SIZE + $self->{data_size},
- content => $buffer
- };
-}
+ # Creating a new file
+ unless ( $bytes_read ) {
+ $self->_write_file_header;
-sub _get_dbm_object {
- my $item = shift;
+ # 1) Create Array/Hash entry
+ my $initial_reference = DBM::Deep::Engine::Sector::Reference->new({
+ engine => $self,
+ type => $obj->_type,
+ });
+ $obj->{base_offset} = $initial_reference->offset;
+ $obj->{staleness} = $initial_reference->staleness;
- my $obj = eval {
- local $SIG{__DIE__};
- if ($item->isa( 'DBM::Deep' )) {
- return $item;
+ $self->storage->flush;
}
- 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;
+ # Reading from an existing file
+ else {
+ $obj->{base_offset} = $bytes_read;
+ my $initial_reference = DBM::Deep::Engine::Sector::Reference->new({
+ engine => $self,
+ offset => $obj->_base_offset,
+ });
+ unless ( $initial_reference ) {
+ DBM::Deep->_throw_error("Corrupted file, no master index record");
}
- 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 $obj;
+
+ unless ($obj->_type eq $initial_reference->type) {
+ DBM::Deep->_throw_error("File type mismatch");
}
- return;
- };
- return $obj if $obj;
+
+ $obj->{staleness} = $initial_reference->staleness;
+ }
}
- return;
+ return 1;
}
-sub _length_needed {
+sub begin_work {
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( $key );
+ my ($obj) = @_;
- if ( $is_dbm_deep && $value->_fileobj eq $self->_fileobj ) {
- return $len + $self->{long_size};
+ if ( $self->trans_id ) {
+ DBM::Deep->_throw_error( "Cannot begin_work within an active transaction" );
}
- 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;
- }
+ my @slots = $self->read_txn_slots;
+ my $found;
+ for my $i ( 0 .. $#slots ) {
+ next if $slots[$i];
- unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
- if ( defined $value ) {
- $len += length( $value );
- }
- return $len;
+ $slots[$i] = 1;
+ $self->set_trans_id( $i + 1 );
+ $found = 1;
+ last;
}
+ unless ( $found ) {
+ DBM::Deep->_throw_error( "Cannot allocate transaction ID" );
+ }
+ $self->write_txn_slots( @slots );
- $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);
- }
+ if ( !$self->trans_id ) {
+ DBM::Deep->_throw_error( "Cannot begin_work - no available transactions" );
}
- return $len;
+ return;
}
-sub add_bucket {
- ##
- # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
- # plain (undigested) key and value.
- ##
+sub rollback {
my $self = shift;
- my ($tag, $md5, $plain_key, $value, $deleted, $orig_key) = @_;
- $deleted ||= 0;
-
- local($/,$\);
-
- # This verifies that only supported values will be stored.
- {
- my $r = Scalar::Util::reftype( $value );
- last if !defined $r;
+ my ($obj) = @_;
- last if $r eq 'HASH';
- last if $r eq 'ARRAY';
+ if ( !$self->trans_id ) {
+ DBM::Deep->_throw_error( "Cannot rollback without an active transaction" );
+ }
- $self->_throw_error(
- "Storage of variables of type '$r' is not supported."
- );
+ # Each entry is the file location for a bucket that has a modification for
+ # this transaction. The entries need to be expunged.
+ foreach my $entry (@{ $self->get_entries } ) {
+ # Remove the entry here
+ my $read_loc = $entry
+ + $self->hash_size
+ + $self->byte_size
+ + $self->byte_size
+ + ($self->trans_id - 1) * ( $self->byte_size + $STALE_SIZE );
+
+ my $data_loc = $self->storage->read_at( $read_loc, $self->byte_size );
+ $data_loc = unpack( $StP{$self->byte_size}, $data_loc );
+ $self->storage->print_at( $read_loc, pack( $StP{$self->byte_size}, 0 ) );
+
+ if ( $data_loc > 1 ) {
+ $self->_load_sector( $data_loc )->free;
+ }
}
- my $location = 0;
- my $result = 2;
+ $self->clear_entries;
- my $fileobj = $self->_fileobj;
+ my @slots = $self->read_txn_slots;
+ $slots[$self->trans_id-1] = 0;
+ $self->write_txn_slots( @slots );
+ $self->inc_txn_staleness_counter( $self->trans_id );
+ $self->set_trans_id( 0 );
- my $actual_length = $self->_length_needed( $value, $plain_key );
+ return 1;
+}
- #ACID - This is a mutation. Must only find the exact transaction
- my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5, 1 );
+sub commit {
+ my $self = shift;
+ my ($obj) = @_;
- my @transactions;
- if ( $fileobj->transaction_id == 0 ) {
- @transactions = $fileobj->current_transactions;
+ if ( !$self->trans_id ) {
+ DBM::Deep->_throw_error( "Cannot commit without an active transaction" );
}
-# $self->_release_space( $size, $subloc );
- # Updating a known md5
-#XXX This needs updating to use _release_space
- if ( $subloc ) {
- $result = 1;
+ foreach my $entry (@{ $self->get_entries } ) {
+ # Overwrite the entry in head with the entry in trans_id
+ my $base = $entry
+ + $self->hash_size
+ + $self->byte_size;
- if ($actual_length <= $size) {
- $location = $subloc;
- }
- else {
- $location = $fileobj->request_space( $actual_length );
+ my $head_loc = $self->storage->read_at( $base, $self->byte_size );
+ $head_loc = unpack( $StP{$self->byte_size}, $head_loc );
- $fileobj->print_at( $tag->{offset} + $offset + $self->{hash_size},
- pack($self->{long_pack}, $location ),
- pack($self->{long_pack}, $actual_length ),
- pack('n n', $fileobj->transaction_id, $deleted ),
- );
- }
- }
- # Adding a new md5
- elsif ( defined $offset ) {
- $location = $fileobj->request_space( $actual_length );
+ my $spot = $base + $self->byte_size + ($self->trans_id - 1) * ( $self->byte_size + $STALE_SIZE );
+ my $trans_loc = $self->storage->read_at(
+ $spot, $self->byte_size,
+ );
- $fileobj->print_at( $tag->{offset} + $offset,
- $md5,
- pack($self->{long_pack}, $location ),
- pack($self->{long_pack}, $actual_length ),
- pack('n n', $fileobj->transaction_id, $deleted ),
+ $self->storage->print_at( $base, $trans_loc );
+ $self->storage->print_at(
+ $spot,
+ pack( $StP{$self->byte_size} . ' ' . $StP{$STALE_SIZE}, (0) x 2 ),
);
- for ( @transactions ) {
- my $tag2 = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
- $fileobj->{transaction_id} = $_;
- $self->add_bucket( $tag2, $md5, '', '', 1, $orig_key );
- $fileobj->{transaction_id} = 0;
+ if ( $head_loc > 1 ) {
+ $self->_load_sector( $head_loc )->free;
}
}
- # If bucket didn't fit into list, split into a new index level
- # split_index() will do the _fileobj->request_space() call
- else {
- $location = $self->split_index( $md5, $tag );
- }
- $self->write_value( $location, $plain_key, $value, $orig_key );
+ $self->clear_entries;
- return $result;
+ my @slots = $self->read_txn_slots;
+ $slots[$self->trans_id-1] = 0;
+ $self->write_txn_slots( @slots );
+ $self->inc_txn_staleness_counter( $self->trans_id );
+ $self->set_trans_id( 0 );
+
+ return 1;
}
-sub write_value {
+sub read_txn_slots {
my $self = shift;
- my ($location, $key, $value, $orig_key) = @_;
-
- my $fileobj = $self->_fileobj;
-
- my $dbm_deep_obj = _get_dbm_object( $value );
- if ( $dbm_deep_obj && $dbm_deep_obj->_fileobj ne $fileobj ) {
- $self->_throw_error( "Cannot cross-reference. Use export() instead" );
- }
-
- ##
- # Write signature based on content type, set content length and write
- # actual value.
- ##
- my $r = Scalar::Util::reftype( $value ) || '';
- if ( $dbm_deep_obj ) {
- $self->write_tag( $location, SIG_INTERNAL,pack($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" );
- }
- $self->write_tag( $location, SIG_HASH, chr(0)x$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( $location, SIG_ARRAY, chr(0)x$self->{index_size} );
- }
- elsif (!defined($value)) {
- $self->write_tag( $location, SIG_NULL, '' );
- }
- else {
- $self->write_tag( $location, SIG_DATA, $value );
- }
+ my $bl = $self->txn_bitfield_len;
+ my $num_bits = $bl * 8;
+ return split '', unpack( 'b'.$num_bits,
+ $self->storage->read_at(
+ $self->trans_loc, $bl,
+ )
+ );
+}
- ##
- # Plain key is stored AFTER value, as keys are typically fetched less often.
- ##
- $fileobj->print_at( undef, pack($self->{data_pack}, length($key)) . $key );
+sub write_txn_slots {
+ my $self = shift;
+ my $num_bits = $self->txn_bitfield_len * 8;
+ $self->storage->print_at( $self->trans_loc,
+ pack( 'b'.$num_bits, join('', @_) ),
+ );
+}
- # Internal references don't care about autobless
- return 1 if $dbm_deep_obj;
+sub get_running_txn_ids {
+ my $self = shift;
+ my @transactions = $self->read_txn_slots;
+ my @trans_ids = map { $_+1} grep { $transactions[$_] } 0 .. $#transactions;
+}
- ##
- # If value is blessed, preserve class name
- ##
- if ( $fileobj->{autobless} ) {
- my $c = Scalar::Util::blessed($value);
- if ( defined $c && !$dbm_deep_obj ) {
- $fileobj->print_at( undef, chr(1), pack($self->{data_pack}, length($c)) . $c );
- }
- else {
- $fileobj->print_at( undef, chr(0) );
- }
- }
+sub get_txn_staleness_counter {
+ my $self = shift;
+ my ($trans_id) = @_;
- ##
- # 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 => $fileobj,
- parent => $self->{obj},
- parent_key => $orig_key,
- };
- %$value = %x;
- }
- elsif ($r eq 'ARRAY') {
- my @x = @$value;
- tie @$value, 'DBM::Deep', {
- base_offset => $location,
- fileobj => $fileobj,
- parent => $self->{obj},
- parent_key => $orig_key,
- };
- @$value = @x;
- }
+ # Hardcode staleness of 0 for the HEAD
+ return 0 unless $trans_id;
- return 1;
+ return unpack( $StP{$STALE_SIZE},
+ $self->storage->read_at(
+ $self->trans_loc + 4 + $STALE_SIZE * ($trans_id - 1),
+ 4,
+ )
+ );
}
-sub split_index {
+sub inc_txn_staleness_counter {
my $self = shift;
- my ($md5, $tag) = @_;
+ my ($trans_id) = @_;
- my $fileobj = $self->_fileobj;
+ # Hardcode staleness of 0 for the HEAD
+ return unless $trans_id;
- my $loc = $fileobj->request_space(
- $self->tag_size( $self->{index_size} ),
+ $self->storage->print_at(
+ $self->trans_loc + 4 + $STALE_SIZE * ($trans_id - 1),
+ pack( $StP{$STALE_SIZE}, $self->get_txn_staleness_counter( $trans_id ) + 1 ),
);
+}
- $fileobj->print_at( $tag->{ref_loc}, pack($self->{long_pack}, $loc) );
+sub get_entries {
+ my $self = shift;
+ return [ keys %{ $self->{entries}{$self->trans_id} ||= {} } ];
+}
- my $index_tag = $self->write_tag(
- $loc, SIG_INDEX,
- chr(0)x$self->{index_size},
- );
+sub add_entry {
+ my $self = shift;
+ my ($trans_id, $loc) = @_;
- my $newtag_loc = $fileobj->request_space(
- $self->tag_size( $self->{bucket_list_size} ),
- );
+ $self->{entries}{$trans_id} ||= {};
+ $self->{entries}{$trans_id}{$loc} = undef;
+}
- my $keys = $tag->{content}
- . $md5 . pack($self->{long_pack}, $newtag_loc)
- . pack($self->{long_pack}, 0) # size
- . pack($self->{long_pack}, 0); # transaction ID
+# If the buckets are being relocated because of a reindexing, the entries
+# mechanism needs to be made aware of it.
+sub reindex_entry {
+ my $self = shift;
+ my ($old_loc, $new_loc) = @_;
+
+ TRANS:
+ while ( my ($trans_id, $locs) = each %{ $self->{entries} } ) {
+ foreach my $orig_loc ( keys %{ $locs } ) {
+ if ( $orig_loc == $old_loc ) {
+ delete $locs->{orig_loc};
+ $locs->{$new_loc} = undef;
+ next TRANS;
+ }
+ }
+ }
+}
- my @newloc = ();
- BUCKET:
- for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
- my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
+sub clear_entries {
+ my $self = shift;
+ delete $self->{entries}{$self->trans_id};
+}
- 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));
+{
+ my $header_fixed = length( SIG_FILE ) + 1 + 4 + 4;
+ my $this_file_version = 2;
- if ($newloc[$num]) {
- my $subkeys = $fileobj->read_at( $newloc[$num], $self->{bucket_list_size} );
+ sub _write_file_header {
+ my $self = shift;
- # This is looking for the first empty spot
- my ($subloc, $offset, $size) = $self->_find_in_buckets(
- { content => $subkeys }, '',
- );
+ my $nt = $self->num_txns;
+ my $bl = $self->txn_bitfield_len;
- $fileobj->print_at( $newloc[$num] + $offset, $key . pack($self->{long_pack}, $old_subloc) );
+ my $header_var = 1 + 1 + 1 + 1 + $bl + $STALE_SIZE * ($nt - 1) + 3 * $self->byte_size;
- next;
- }
+ my $loc = $self->storage->request_space( $header_fixed + $header_var );
- my $loc = $fileobj->request_space(
- $self->tag_size( $self->{bucket_list_size} ),
- );
+ $self->storage->print_at( $loc,
+ SIG_FILE,
+ SIG_HEADER,
+ pack('N', $this_file_version), # At this point, we're at 9 bytes
+ pack('N', $header_var), # header size
+ # --- Above is $header_fixed. Below is $header_var
+ pack('C', $self->byte_size),
- $fileobj->print_at(
- $index_tag->{offset} + ($num * $self->{long_size}),
- pack($self->{long_pack}, $loc),
- );
+ # These shenanigans are to allow a 256 within a C
+ pack('C', $self->max_buckets - 1),
+ pack('C', $self->data_sector_size - 1),
- my $blist_tag = $self->write_tag(
- $loc, SIG_BLIST,
- chr(0)x$self->{bucket_list_size},
+ pack('C', $nt),
+ pack('C' . $bl, 0 ), # Transaction activeness bitfield
+ pack($StP{$STALE_SIZE}.($nt-1), 0 x ($nt-1) ), # Transaction staleness counters
+ pack($StP{$self->byte_size}, 0), # Start of free chain (blist size)
+ pack($StP{$self->byte_size}, 0), # Start of free chain (data size)
+ pack($StP{$self->byte_size}, 0), # Start of free chain (index size)
);
- $fileobj->print_at( $blist_tag->{offset}, $key . pack($self->{long_pack}, $old_subloc) );
+ #XXX Set these less fragilely
+ $self->set_trans_loc( $header_fixed + 4 );
+ $self->set_chains_loc( $header_fixed + 4 + $bl + $STALE_SIZE * ($nt-1) );
- $newloc[$num] = $blist_tag->{offset};
+ return;
}
- $self->_release_space(
- $self->tag_size( $self->{bucket_list_size} ),
- $tag->{offset} - SIG_SIZE - $self->{data_size},
- );
-
- return $newtag_loc;
-}
+ sub _read_file_header {
+ my $self = shift;
-sub read_from_loc {
- my $self = shift;
- my ($subloc, $orig_key) = @_;
+ my $buffer = $self->storage->read_at( 0, $header_fixed );
+ return unless length($buffer);
- my $fileobj = $self->_fileobj;
+ my ($file_signature, $sig_header, $file_version, $size) = unpack(
+ 'A4 A N N', $buffer
+ );
- ##
- # Found match -- seek to offset and read signature
- ##
- my $signature = $fileobj->read_at( $subloc, SIG_SIZE );
+ unless ( $file_signature eq SIG_FILE ) {
+ $self->storage->close;
+ DBM::Deep->_throw_error( "Signature not found -- file is not a Deep DB" );
+ }
- ##
- # If value is a hash or array, return new DBM::Deep object with correct offset
- ##
- if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
- my $new_obj = DBM::Deep->new({
- type => $signature,
- base_offset => $subloc,
- fileobj => $self->_fileobj,
- parent => $self->{obj},
- parent_key => $orig_key,
- });
+ unless ( $sig_header eq SIG_HEADER ) {
+ $self->storage->close;
+ DBM::Deep->_throw_error( "Pre-1.00 file version found" );
+ }
- if ($new_obj->_fileobj->{autobless}) {
- ##
- # Skip over value and plain key to see if object needs
- # to be re-blessed
- ##
- $fileobj->increment_pointer( $self->{data_size} + $self->{index_size} );
-
- my $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
- if ($size) { $fileobj->increment_pointer( $size ); }
-
- my $bless_bit = $fileobj->read_at( undef, 1 );
- if (ord($bless_bit)) {
- ##
- # Yes, object needs to be re-blessed
- ##
- my $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
-
- my $class_name;
- if ($size) { $class_name = $fileobj->read_at( undef, $size ); }
- if (defined $class_name) { $new_obj = bless( $new_obj, $class_name ); }
- }
+ unless ( $file_version == $this_file_version ) {
+ $self->storage->close;
+ DBM::Deep->_throw_error(
+ "Wrong file version found - " . $file_version .
+ " - expected " . $this_file_version
+ );
}
- return $new_obj;
- }
- elsif ( $signature eq SIG_INTERNAL ) {
- my $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
+ my $buffer2 = $self->storage->read_at( undef, $size );
+ my @values = unpack( 'C C C C', $buffer2 );
- if ( $size ) {
- my $new_loc = $fileobj->read_at( undef, $size );
- $new_loc = unpack( $self->{long_pack}, $new_loc );
- return $self->read_from_loc( $new_loc, $orig_key );
+ if ( @values != 4 || grep { !defined } @values ) {
+ $self->storage->close;
+ DBM::Deep->_throw_error("Corrupted file - bad header");
}
- else {
- return;
- }
- }
- ##
- # Otherwise return actual value
- ##
- elsif ( $signature eq SIG_DATA ) {
- my $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
- my $value = '';
- if ($size) { $value = $fileobj->read_at( undef, $size ); }
- return $value;
- }
+ #XXX Add warnings if values weren't set right
+ @{$self}{qw(byte_size max_buckets data_sector_size num_txns)} = @values;
- ##
- # Key exists, but content is null
- ##
- return;
-}
+ # These shenangians are to allow a 256 within a C
+ $self->{max_buckets} += 1;
+ $self->{data_sector_size} += 1;
-sub get_bucket_value {
- ##
- # Fetch single value given tag and MD5 digested key.
- ##
- my $self = shift;
- my ($tag, $md5, $orig_key) = @_;
+ my $bl = $self->txn_bitfield_len;
- #ACID - This is a read. Can find exact or HEAD
- my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5 );
- if ( $subloc && !$is_deleted ) {
- return $self->read_from_loc( $subloc, $orig_key );
+ my $header_var = scalar(@values) + $bl + $STALE_SIZE * ($self->num_txns - 1) + 3 * $self->byte_size;
+ unless ( $size == $header_var ) {
+ $self->storage->close;
+ DBM::Deep->_throw_error( "Unexpected size found ($size <-> $header_var)." );
+ }
+
+ $self->set_trans_loc( $header_fixed + scalar(@values) );
+ $self->set_chains_loc( $header_fixed + scalar(@values) + $bl + $STALE_SIZE * ($self->num_txns - 1) );
+
+ return length($buffer) + length($buffer2);
}
- return;
}
-sub delete_bucket {
- ##
- # Delete single key/value pair given tag and MD5 digested key.
- ##
+sub _load_sector {
my $self = shift;
- my ($tag, $md5, $orig_key) = @_;
+ my ($offset) = @_;
- #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() for the value and anything below
- if ( $subloc ) {
- $self->_fileobj->print_at(
- $tag->{offset} + $offset,
- substr($tag->{content}, $offset + $self->{bucket_size} ),
- chr(0) x $self->{bucket_size},
- );
+ # Add a catch for offset of 0 or 1
+ return if $offset <= 1;
- return 1;
+ my $type = $self->storage->read_at( $offset, 1 );
+ return if $type eq chr(0);
+
+ if ( $type eq $self->SIG_ARRAY || $type eq $self->SIG_HASH ) {
+ return DBM::Deep::Engine::Sector::Reference->new({
+ engine => $self,
+ type => $type,
+ offset => $offset,
+ });
}
- return;
+ # XXX Don't we need key_md5 here?
+ elsif ( $type eq $self->SIG_BLIST ) {
+ return DBM::Deep::Engine::Sector::BucketList->new({
+ engine => $self,
+ type => $type,
+ offset => $offset,
+ });
+ }
+ elsif ( $type eq $self->SIG_INDEX ) {
+ return DBM::Deep::Engine::Sector::Index->new({
+ engine => $self,
+ type => $type,
+ offset => $offset,
+ });
+ }
+ elsif ( $type eq $self->SIG_NULL ) {
+ return DBM::Deep::Engine::Sector::Null->new({
+ engine => $self,
+ type => $type,
+ offset => $offset,
+ });
+ }
+ elsif ( $type eq $self->SIG_DATA ) {
+ return DBM::Deep::Engine::Sector::Scalar->new({
+ engine => $self,
+ type => $type,
+ offset => $offset,
+ });
+ }
+ # This was deleted from under us, so just return and let the caller figure it out.
+ elsif ( $type eq $self->SIG_FREE ) {
+ return;
+ }
+
+ DBM::Deep->_throw_error( "'$offset': Don't know what to do with type '$type'" );
}
-sub bucket_exists {
- ##
- # Check existence of single key given tag and MD5 digested key.
- ##
+sub _apply_digest {
my $self = shift;
- my ($tag, $md5) = @_;
-
- #ACID - This is a read. Can find exact or HEAD
- my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
- return ($subloc && !$is_deleted) && 1;
+ return $self->{digest}->(@_);
}
-sub find_bucket_list {
- ##
- # Locate offset for bucket list, given digested key
- ##
+sub _add_free_blist_sector { shift->_add_free_sector( 0, @_ ) }
+sub _add_free_data_sector { shift->_add_free_sector( 1, @_ ) }
+sub _add_free_index_sector { shift->_add_free_sector( 2, @_ ) }
+
+sub _add_free_sector {
my $self = shift;
- my ($offset, $md5, $args) = @_;
- $args = {} unless $args;
+ my ($multiple, $offset, $size) = @_;
- local($/,$\);
+ my $chains_offset = $multiple * $self->byte_size;
- ##
- # Locate offset for bucket list using digest index system
- ##
- my $tag = $self->load_tag( $offset )
- or $self->_throw_error( "INTERNAL ERROR - Cannot find tag" );
+ my $storage = $self->storage;
- my $ch = 0;
- while ($tag->{signature} ne SIG_BLIST) {
- my $num = ord substr($md5, $ch, 1);
+ # Increment staleness.
+ # XXX Can this increment+modulo be done by "&= 0x1" ?
+ my $staleness = unpack( $StP{$STALE_SIZE}, $storage->read_at( $offset + SIG_SIZE, $STALE_SIZE ) );
+ $staleness = ($staleness + 1 ) % ( 2 ** ( 8 * $STALE_SIZE ) );
+ $storage->print_at( $offset + SIG_SIZE, pack( $StP{$STALE_SIZE}, $staleness ) );
- my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
- $tag = $self->index_lookup( $tag, $num );
+ my $old_head = $storage->read_at( $self->chains_loc + $chains_offset, $self->byte_size );
- if (!$tag) {
- return if !$args->{create};
+ $storage->print_at( $self->chains_loc + $chains_offset,
+ pack( $StP{$self->byte_size}, $offset ),
+ );
- my $loc = $self->_fileobj->request_space(
- $self->tag_size( $self->{bucket_list_size} ),
- );
+ # Record the old head in the new sector after the signature and staleness counter
+ $storage->print_at( $offset + SIG_SIZE + $STALE_SIZE, $old_head );
+}
- $self->_fileobj->print_at( $ref_loc, pack($self->{long_pack}, $loc) );
+sub _request_blist_sector { shift->_request_sector( 0, @_ ) }
+sub _request_data_sector { shift->_request_sector( 1, @_ ) }
+sub _request_index_sector { shift->_request_sector( 2, @_ ) }
- $tag = $self->write_tag(
- $loc, SIG_BLIST,
- chr(0)x$self->{bucket_list_size},
- );
+sub _request_sector {
+ my $self = shift;
+ my ($multiple, $size) = @_;
- $tag->{ref_loc} = $ref_loc;
- $tag->{ch} = $ch;
+ my $chains_offset = $multiple * $self->byte_size;
- last;
- }
+ my $old_head = $self->storage->read_at( $self->chains_loc + $chains_offset, $self->byte_size );
+ my $loc = unpack( $StP{$self->byte_size}, $old_head );
- $tag->{ch} = $ch++;
- $tag->{ref_loc} = $ref_loc;
- }
+ # We don't have any free sectors of the right size, so allocate a new one.
+ unless ( $loc ) {
+ my $offset = $self->storage->request_space( $size );
- return $tag;
-}
+ # Zero out the new sector. This also guarantees correct increases
+ # in the filesize.
+ $self->storage->print_at( $offset, chr(0) x $size );
-sub index_lookup {
- ##
- # Given index tag, lookup single entry in index and return .
- ##
- my $self = shift;
- my ($tag, $index) = @_;
+ return $offset;
+ }
- my $location = unpack(
- $self->{long_pack},
- substr(
- $tag->{content},
- $index * $self->{long_size},
- $self->{long_size},
- ),
+ # Read the new head after the signature and the staleness counter
+ my $new_head = $self->storage->read_at( $loc + SIG_SIZE + $STALE_SIZE, $self->byte_size );
+ $self->storage->print_at( $self->chains_loc + $chains_offset, $new_head );
+ $self->storage->print_at(
+ $loc + SIG_SIZE + $STALE_SIZE,
+ pack( $StP{$self->byte_size}, 0 ),
);
- if (!$location) { return; }
-
- return $self->load_tag( $location );
+ return $loc;
}
-sub traverse_index {
- ##
- # Scan index and recursively step into deeper levels, looking for next key.
- ##
+################################################################################
+
+sub storage { $_[0]{storage} }
+sub byte_size { $_[0]{byte_size} }
+sub hash_size { $_[0]{hash_size} }
+sub hash_chars { $_[0]{hash_chars} }
+sub num_txns { $_[0]{num_txns} }
+sub max_buckets { $_[0]{max_buckets} }
+sub blank_md5 { chr(0) x $_[0]->hash_size }
+sub data_sector_size { $_[0]{data_sector_size} }
+
+# This is a calculated value
+sub txn_bitfield_len {
my $self = shift;
- my ($obj, $offset, $ch, $force_return_next) = @_;
+ unless ( exists $self->{txn_bitfield_len} ) {
+ my $temp = ($self->num_txns) / 8;
+ if ( $temp > int( $temp ) ) {
+ $temp = int( $temp ) + 1;
+ }
+ $self->{txn_bitfield_len} = $temp;
+ }
+ return $self->{txn_bitfield_len};
+}
- my $tag = $self->load_tag( $offset );
+sub trans_id { $_[0]{trans_id} }
+sub set_trans_id { $_[0]{trans_id} = $_[1] }
- if ($tag->{signature} ne SIG_BLIST) {
- my $content = $tag->{content};
- my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
+sub trans_loc { $_[0]{trans_loc} }
+sub set_trans_loc { $_[0]{trans_loc} = $_[1] }
- for (my $idx = $start; $idx < (2**8); $idx++) {
- my $subloc = unpack(
- $self->{long_pack},
- substr(
- $content,
- $idx * $self->{long_size},
- $self->{long_size},
- ),
- );
+sub chains_loc { $_[0]{chains_loc} }
+sub set_chains_loc { $_[0]{chains_loc} = $_[1] }
- if ($subloc) {
- my $result = $self->traverse_index(
- $obj, $subloc, $ch + 1, $force_return_next,
- );
+################################################################################
- if (defined($result)) { return $result; }
- }
- } # index loop
+package DBM::Deep::Iterator;
- $obj->{return_next} = 1;
- } # tag is an index
+sub new {
+ my $class = shift;
+ my ($args) = @_;
- else {
- my $keys = $tag->{content};
- if ($force_return_next) { $obj->{return_next} = 1; }
-
- ##
- # Iterate through buckets, looking for a key match
- ##
- for (my $i = 0; $i < $self->{max_buckets}; $i++) {
- my ($key, $subloc) = $self->_get_key_subloc( $keys, $i );
-
- # End of bucket list -- return to outer loop
- if (!$subloc) {
- $obj->{return_next} = 1;
- last;
- }
- # Located previous key -- return next one found
- elsif ($key eq $obj->{prev_md5}) {
- $obj->{return_next} = 1;
- next;
- }
- # Seek to bucket location and skip over signature
- elsif ($obj->{return_next}) {
- my $fileobj = $self->_fileobj;
+ my $self = bless {
+ breadcrumbs => [],
+ engine => $args->{engine},
+ base_offset => $args->{base_offset},
+ }, $class;
- # Skip over value to get to plain key
- my $sig = $fileobj->read_at( $subloc, SIG_SIZE );
+ Scalar::Util::weaken( $self->{engine} );
- my $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
- if ($size) { $fileobj->increment_pointer( $size ); }
+ return $self;
+}
- # Read in plain key and return as scalar
- $size = $fileobj->read_at( undef, $self->{data_size} );
- $size = unpack($self->{data_pack}, $size);
- my $plain_key;
- if ($size) { $plain_key = $fileobj->read_at( undef, $size); }
+sub reset { $_[0]{breadcrumbs} = [] }
- return $plain_key;
- }
- }
+sub get_sector_iterator {
+ my $self = shift;
+ my ($loc) = @_;
- $obj->{return_next} = 1;
- } # tag is a bucket list
+ my $sector = $self->{engine}->_load_sector( $loc )
+ or return;
- return;
+ if ( $sector->isa( 'DBM::Deep::Engine::Sector::Index' ) ) {
+ return DBM::Deep::Iterator::Index->new({
+ iterator => $self,
+ sector => $sector,
+ });
+ }
+ elsif ( $sector->isa( 'DBM::Deep::Engine::Sector::BucketList' ) ) {
+ return DBM::Deep::Iterator::BucketList->new({
+ iterator => $self,
+ sector => $sector,
+ });
+ }
+
+ DBM::Deep->_throw_error( "get_sector_iterator(): Why did $loc make a $sector?" );
}
sub get_next_key {
- ##
- # Locate next key, given digested previous one
- ##
my $self = shift;
my ($obj) = @_;
- $obj->{prev_md5} = $_[1] ? $_[1] : undef;
- $obj->{return_next} = 0;
-
- ##
- # If the previous key was not specifed, start at the top and
- # return the first one found.
- ##
- if (!$obj->{prev_md5}) {
- $obj->{prev_md5} = chr(0) x $self->{hash_size};
- $obj->{return_next} = 1;
- }
+ my $crumbs = $self->{breadcrumbs};
+ my $e = $self->{engine};
- return $self->traverse_index( $obj, $obj->_base_offset, 0 );
-}
+ unless ( @$crumbs ) {
+ # This will be a Reference sector
+ my $sector = $e->_load_sector( $self->{base_offset} )
+ # If no sector is found, thist must have been deleted from under us.
+ or return;
-# Utilities
+ if ( $sector->staleness != $obj->_staleness ) {
+ return;
+ }
+
+ my $loc = $sector->get_blist_loc
+ or return;
+
+ push @$crumbs, $self->get_sector_iterator( $loc );
+ }
+
+ FIND_NEXT_KEY: {
+ # We're at the end.
+ unless ( @$crumbs ) {
+ $self->reset;
+ return;
+ }
+
+ my $iterator = $crumbs->[-1];
+
+ # This level is done.
+ if ( $iterator->at_end ) {
+ pop @$crumbs;
+ redo FIND_NEXT_KEY;
+ }
+
+ if ( $iterator->isa( 'DBM::Deep::Iterator::Index' ) ) {
+ # If we don't have any more, it will be caught at the
+ # prior check.
+ if ( my $next = $iterator->get_next_iterator ) {
+ push @$crumbs, $next;
+ }
+ redo FIND_NEXT_KEY;
+ }
+
+ unless ( $iterator->isa( 'DBM::Deep::Iterator::BucketList' ) ) {
+ DBM::Deep->_throw_error(
+ "Should have a bucketlist iterator here - instead have $iterator"
+ );
+ }
+
+ # At this point, we have a BucketList iterator
+ my $key = $iterator->get_next_key;
+ if ( defined $key ) {
+ return $key;
+ }
+ #XXX else { $iterator->set_to_end() } ?
+
+ # We hit the end of the bucketlist iterator, so redo
+ redo FIND_NEXT_KEY;
+ }
+
+ DBM::Deep->_throw_error( "get_next_key(): How did we get here?" );
+}
+
+package DBM::Deep::Iterator::Index;
+
+sub new {
+ my $self = bless $_[1] => $_[0];
+ $self->{curr_index} = 0;
+ return $self;
+}
+
+sub at_end {
+ my $self = shift;
+ return $self->{curr_index} >= $self->{iterator}{engine}->hash_chars;
+}
+
+sub get_next_iterator {
+ my $self = shift;
+
+ my $loc;
+ while ( !$loc ) {
+ return if $self->at_end;
+ $loc = $self->{sector}->get_entry( $self->{curr_index}++ );
+ }
+
+ return $self->{iterator}->get_sector_iterator( $loc );
+}
+
+package DBM::Deep::Iterator::BucketList;
+
+sub new {
+ my $self = bless $_[1] => $_[0];
+ $self->{curr_index} = 0;
+ return $self;
+}
+
+sub at_end {
+ my $self = shift;
+ return $self->{curr_index} >= $self->{iterator}{engine}->max_buckets;
+}
+
+sub get_next_key {
+ my $self = shift;
+
+ return if $self->at_end;
+
+ my $idx = $self->{curr_index}++;
+
+ my $data_loc = $self->{sector}->get_data_location_for({
+ allow_head => 1,
+ idx => $idx,
+ }) or return;
+
+ #XXX Do we want to add corruption checks here?
+ return $self->{sector}->get_key_for( $idx )->data;
+}
+
+package DBM::Deep::Engine::Sector;
+
+sub new {
+ my $self = bless $_[1], $_[0];
+ Scalar::Util::weaken( $self->{engine} );
+ $self->_init;
+ return $self;
+}
+
+#sub _init {}
+#sub clone { DBM::Deep->_throw_error( "Must be implemented in the child class" ); }
+
+sub engine { $_[0]{engine} }
+sub offset { $_[0]{offset} }
+sub type { $_[0]{type} }
+
+sub base_size {
+ my $self = shift;
+ return $self->engine->SIG_SIZE + $STALE_SIZE;
+}
+
+sub free {
+ my $self = shift;
+
+ my $e = $self->engine;
+
+ $e->storage->print_at( $self->offset, $e->SIG_FREE );
+ # Skip staleness counter
+ $e->storage->print_at( $self->offset + $self->base_size,
+ chr(0) x ($self->size - $self->base_size),
+ );
+
+ my $free_meth = $self->free_meth;
+ $e->$free_meth( $self->offset, $self->size );
+
+ return;
+}
+
+package DBM::Deep::Engine::Sector::Data;
+
+our @ISA = qw( DBM::Deep::Engine::Sector );
+
+# This is in bytes
+sub size { $_[0]{engine}->data_sector_size }
+sub free_meth { return '_add_free_data_sector' }
+
+sub clone {
+ my $self = shift;
+ return ref($self)->new({
+ engine => $self->engine,
+ type => $self->type,
+ data => $self->data,
+ });
+}
+
+package DBM::Deep::Engine::Sector::Scalar;
+
+our @ISA = qw( DBM::Deep::Engine::Sector::Data );
+
+sub free {
+ my $self = shift;
+
+ my $chain_loc = $self->chain_loc;
+
+ $self->SUPER::free();
+
+ if ( $chain_loc ) {
+ $self->engine->_load_sector( $chain_loc )->free;
+ }
+
+ return;
+}
+
+sub type { $_[0]{engine}->SIG_DATA }
+sub _init {
+ my $self = shift;
+
+ my $engine = $self->engine;
+
+ unless ( $self->offset ) {
+ my $data_section = $self->size - $self->base_size - $engine->byte_size - 1;
+
+ $self->{offset} = $engine->_request_data_sector( $self->size );
+
+ my $data = delete $self->{data};
+ my $dlen = length $data;
+ my $continue = 1;
+ my $curr_offset = $self->offset;
+ while ( $continue ) {
+
+ my $next_offset = 0;
+
+ my ($leftover, $this_len, $chunk);
+ if ( $dlen > $data_section ) {
+ $leftover = 0;
+ $this_len = $data_section;
+ $chunk = substr( $data, 0, $this_len );
+
+ $dlen -= $data_section;
+ $next_offset = $engine->_request_data_sector( $self->size );
+ $data = substr( $data, $this_len );
+ }
+ else {
+ $leftover = $data_section - $dlen;
+ $this_len = $dlen;
+ $chunk = $data;
+
+ $continue = 0;
+ }
+
+ $engine->storage->print_at( $curr_offset, $self->type ); # Sector type
+ # Skip staleness
+ $engine->storage->print_at( $curr_offset + $self->base_size,
+ pack( $StP{$engine->byte_size}, $next_offset ), # Chain loc
+ pack( $StP{1}, $this_len ), # Data length
+ $chunk, # Data to be stored in this sector
+ chr(0) x $leftover, # Zero-fill the rest
+ );
+
+ $curr_offset = $next_offset;
+ }
+
+ return;
+ }
+}
-sub _get_key_subloc {
+sub data_length {
my $self = shift;
- my ($keys, $idx) = @_;
- my ($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}),
- $self->{bucket_size},
+ my $buffer = $self->engine->storage->read_at(
+ $self->offset + $self->base_size + $self->engine->byte_size, 1
+ );
+
+ return unpack( $StP{1}, $buffer );
+}
+
+sub chain_loc {
+ my $self = shift;
+ return unpack(
+ $StP{$self->engine->byte_size},
+ $self->engine->storage->read_at(
+ $self->offset + $self->base_size,
+ $self->engine->byte_size,
),
);
+}
+
+sub data {
+ my $self = shift;
+
+ my $data;
+ while ( 1 ) {
+ my $chain_loc = $self->chain_loc;
+
+ $data .= $self->engine->storage->read_at(
+ $self->offset + $self->base_size + $self->engine->byte_size + 1, $self->data_length,
+ );
+
+ last unless $chain_loc;
- return ($key, $subloc, $size, $transaction_id, $is_deleted);
+ $self = $self->engine->_load_sector( $chain_loc );
+ }
+
+ return $data;
}
-sub _find_in_buckets {
+package DBM::Deep::Engine::Sector::Null;
+
+our @ISA = qw( DBM::Deep::Engine::Sector::Data );
+
+sub type { $_[0]{engine}->SIG_NULL }
+sub data_length { 0 }
+sub data { return }
+
+sub _init {
my $self = shift;
- my ($tag, $md5, $exact) = @_;
- my $trans_id = $self->_fileobj->transaction_id;
+ my $engine = $self->engine;
- my @zero;
+ unless ( $self->offset ) {
+ my $leftover = $self->size - $self->base_size - 1 * $engine->byte_size - 1;
- 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,
+ $self->{offset} = $engine->_request_data_sector( $self->size );
+ $engine->storage->print_at( $self->offset, $self->type ); # Sector type
+ # Skip staleness counter
+ $engine->storage->print_at( $self->offset + $self->base_size,
+ pack( $StP{$engine->byte_size}, 0 ), # Chain loc
+ pack( $StP{1}, $self->data_length ), # Data length
+ chr(0) x $leftover, # Zero-fill the rest
);
- my @rv = ($subloc, $i * $self->{bucket_size}, $size, $is_deleted);
+ return;
+ }
+}
+
+package DBM::Deep::Engine::Sector::Reference;
+
+our @ISA = qw( DBM::Deep::Engine::Sector::Data );
+
+sub _init {
+ my $self = shift;
+
+ my $e = $self->engine;
+
+ unless ( $self->offset ) {
+ my $classname = Scalar::Util::blessed( delete $self->{data} );
+ my $leftover = $self->size - $self->base_size - 2 * $e->byte_size;
+
+ my $class_offset = 0;
+ if ( defined $classname ) {
+ my $class_sector = DBM::Deep::Engine::Sector::Scalar->new({
+ engine => $e,
+ data => $classname,
+ });
+ $class_offset = $class_sector->offset;
+ }
+
+ $self->{offset} = $e->_request_data_sector( $self->size );
+ $e->storage->print_at( $self->offset, $self->type ); # Sector type
+ # Skip staleness counter
+ $e->storage->print_at( $self->offset + $self->base_size,
+ pack( $StP{$e->byte_size}, 0 ), # Index/BList loc
+ pack( $StP{$e->byte_size}, $class_offset ), # Classname loc
+ chr(0) x $leftover, # Zero-fill the rest
+ );
+ }
+ else {
+ $self->{type} = $e->storage->read_at( $self->offset, 1 );
+ }
+
+ $self->{staleness} = unpack(
+ $StP{$STALE_SIZE},
+ $e->storage->read_at( $self->offset + $e->SIG_SIZE, $STALE_SIZE ),
+ );
+
+ return;
+}
+
+sub free {
+ my $self = shift;
+
+ my $blist_loc = $self->get_blist_loc;
+ $self->engine->_load_sector( $blist_loc )->free if $blist_loc;
+
+ my $class_loc = $self->get_class_offset;
+ $self->engine->_load_sector( $class_loc )->free if $class_loc;
+
+ $self->SUPER::free();
+}
+
+sub staleness { $_[0]{staleness} }
+
+sub get_data_for {
+ my $self = shift;
+ my ($args) = @_;
+
+ # Assume that the head is not allowed unless otherwise specified.
+ $args->{allow_head} = 0 unless exists $args->{allow_head};
+
+ # Assume we don't create a new blist location unless otherwise specified.
+ $args->{create} = 0 unless exists $args->{create};
+
+ my $blist = $self->get_bucket_list({
+ key_md5 => $args->{key_md5},
+ key => $args->{key},
+ create => $args->{create},
+ });
+ return unless $blist && $blist->{found};
- unless ( $subloc ) {
- if ( !$exact && @zero and $trans_id ) {
- @rv = ($zero[2], $zero[0] * $self->{bucket_size},$zero[3],$is_deleted);
+ # At this point, $blist knows where the md5 is. What it -doesn't- know yet
+ # is whether or not this transaction has this key. That's part of the next
+ # function call.
+ my $location = $blist->get_data_location_for({
+ allow_head => $args->{allow_head},
+ }) or return;
+
+ return $self->engine->_load_sector( $location );
+}
+
+sub write_data {
+ my $self = shift;
+ my ($args) = @_;
+
+ my $blist = $self->get_bucket_list({
+ key_md5 => $args->{key_md5},
+ key => $args->{key},
+ create => 1,
+ }) or DBM::Deep->_throw_error( "How did write_data fail (no blist)?!" );
+
+ # Handle any transactional bookkeeping.
+ if ( $self->engine->trans_id ) {
+ if ( ! $blist->has_md5 ) {
+ $blist->mark_deleted({
+ trans_id => 0,
+ });
+ }
+ }
+ else {
+ my @trans_ids = $self->engine->get_running_txn_ids;
+ if ( $blist->has_md5 ) {
+ if ( @trans_ids ) {
+ my $old_value = $blist->get_data_for;
+ foreach my $other_trans_id ( @trans_ids ) {
+ next if $blist->get_data_location_for({
+ trans_id => $other_trans_id,
+ allow_head => 0,
+ });
+ $blist->write_md5({
+ trans_id => $other_trans_id,
+ key => $args->{key},
+ key_md5 => $args->{key_md5},
+ value => $old_value->clone,
+ });
+ }
}
- return @rv;
}
+ else {
+ if ( @trans_ids ) {
+ foreach my $other_trans_id ( @trans_ids ) {
+ #XXX This doesn't seem to possible to ever happen . . .
+ next if $blist->get_data_location_for({ trans_id => $other_trans_id, allow_head => 0 });
+ $blist->mark_deleted({
+ trans_id => $other_trans_id,
+ });
+ }
+ }
+ }
+ }
- next BUCKET if $key ne $md5;
+ #XXX Is this safe to do transactionally?
+ # Free the place we're about to write to.
+ if ( $blist->get_data_location_for({ allow_head => 0 }) ) {
+ $blist->get_data_for({ allow_head => 0 })->free;
+ }
- # Save off the HEAD in case we need it.
- @zero = ($i,$key,$subloc,$size,$transaction_id,$is_deleted) if $transaction_id == 0;
+ $blist->write_md5({
+ key => $args->{key},
+ key_md5 => $args->{key_md5},
+ value => $args->{value},
+ });
+}
- next BUCKET if $transaction_id != $trans_id;
+sub delete_key {
+ my $self = shift;
+ my ($args) = @_;
- return @rv;
+ # XXX What should happen if this fails?
+ my $blist = $self->get_bucket_list({
+ key_md5 => $args->{key_md5},
+ }) or DBM::Deep->_throw_error( "How did delete_key fail (no blist)?!" );
+
+ # Save the location so that we can free the data
+ my $location = $blist->get_data_location_for({
+ allow_head => 0,
+ });
+ my $old_value = $location && $self->engine->_load_sector( $location );
+
+ my @trans_ids = $self->engine->get_running_txn_ids;
+
+ if ( $self->engine->trans_id == 0 ) {
+ if ( @trans_ids ) {
+ foreach my $other_trans_id ( @trans_ids ) {
+ next if $blist->get_data_location_for({ trans_id => $other_trans_id, allow_head => 0 });
+ $blist->write_md5({
+ trans_id => $other_trans_id,
+ key => $args->{key},
+ key_md5 => $args->{key_md5},
+ value => $old_value->clone,
+ });
+ }
+ }
}
- return;
+ my $data;
+ if ( @trans_ids ) {
+ $blist->mark_deleted( $args );
+
+ if ( $old_value ) {
+ $data = $old_value->data;
+ $old_value->free;
+ }
+ }
+ else {
+ $data = $blist->delete_md5( $args );
+ }
+
+ return $data;
}
-sub _release_space {
+sub get_blist_loc {
my $self = shift;
- my ($size, $loc) = @_;
- my $next_loc = 0;
+ my $e = $self->engine;
+ my $blist_loc = $e->storage->read_at( $self->offset + $self->base_size, $e->byte_size );
+ return unpack( $StP{$e->byte_size}, $blist_loc );
+}
+
+sub get_bucket_list {
+ my $self = shift;
+ my ($args) = @_;
+ $args ||= {};
+
+ # XXX Add in check here for recycling?
+
+ my $engine = $self->engine;
+
+ my $blist_loc = $self->get_blist_loc;
+
+ # There's no index or blist yet
+ unless ( $blist_loc ) {
+ return unless $args->{create};
+
+ my $blist = DBM::Deep::Engine::Sector::BucketList->new({
+ engine => $engine,
+ key_md5 => $args->{key_md5},
+ });
+
+ $engine->storage->print_at( $self->offset + $self->base_size,
+ pack( $StP{$engine->byte_size}, $blist->offset ),
+ );
+
+ return $blist;
+ }
+
+ my $sector = $engine->_load_sector( $blist_loc )
+ or DBM::Deep->_throw_error( "Cannot read sector at $blist_loc in get_bucket_list()" );
+ my $i = 0;
+ my $last_sector = undef;
+ while ( $sector->isa( 'DBM::Deep::Engine::Sector::Index' ) ) {
+ $blist_loc = $sector->get_entry( ord( substr( $args->{key_md5}, $i++, 1 ) ) );
+ $last_sector = $sector;
+ if ( $blist_loc ) {
+ $sector = $engine->_load_sector( $blist_loc )
+ or DBM::Deep->_throw_error( "Cannot read sector at $blist_loc in get_bucket_list()" );
+ }
+ else {
+ $sector = undef;
+ last;
+ }
+ }
+
+ # This means we went through the Index sector(s) and found an empty slot
+ unless ( $sector ) {
+ return unless $args->{create};
+
+ DBM::Deep->_throw_error( "No last_sector when attempting to build a new entry" )
+ unless $last_sector;
+
+ my $blist = DBM::Deep::Engine::Sector::BucketList->new({
+ engine => $engine,
+ key_md5 => $args->{key_md5},
+ });
+
+ $last_sector->set_entry( ord( substr( $args->{key_md5}, $i - 1, 1 ) ) => $blist->offset );
+
+ return $blist;
+ }
- $self->_fileobj->print_at( $loc,
- SIG_FREE,
- pack($self->{long_pack}, $size ),
- pack($self->{long_pack}, $next_loc ),
+ $sector->find_md5( $args->{key_md5} );
+
+ # See whether or not we need to reindex the bucketlist
+ if ( !$sector->has_md5 && $args->{create} && $sector->{idx} == -1 ) {
+ my $new_index = DBM::Deep::Engine::Sector::Index->new({
+ engine => $engine,
+ });
+
+ my %blist_cache;
+ #XXX q.v. the comments for this function.
+ foreach my $entry ( $sector->chopped_up ) {
+ my ($spot, $md5) = @{$entry};
+ my $idx = ord( substr( $md5, $i, 1 ) );
+
+ # XXX This is inefficient
+ my $blist = $blist_cache{$idx}
+ ||= DBM::Deep::Engine::Sector::BucketList->new({
+ engine => $engine,
+ });
+
+ $new_index->set_entry( $idx => $blist->offset );
+
+ my $new_spot = $blist->write_at_next_open( $md5 );
+ $engine->reindex_entry( $spot => $new_spot );
+ }
+
+ # Handle the new item separately.
+ {
+ my $idx = ord( substr( $args->{key_md5}, $i, 1 ) );
+ my $blist = $blist_cache{$idx}
+ ||= DBM::Deep::Engine::Sector::BucketList->new({
+ engine => $engine,
+ });
+
+ $new_index->set_entry( $idx => $blist->offset );
+
+ #XXX THIS IS HACKY!
+ $blist->find_md5( $args->{key_md5} );
+ $blist->write_md5({
+ key => $args->{key},
+ key_md5 => $args->{key_md5},
+ value => DBM::Deep::Engine::Sector::Null->new({
+ engine => $engine,
+ data => undef,
+ }),
+ });
+ }
+
+ if ( $last_sector ) {
+ $last_sector->set_entry(
+ ord( substr( $args->{key_md5}, $i - 1, 1 ) ),
+ $new_index->offset,
+ );
+ } else {
+ $engine->storage->print_at( $self->offset + $self->base_size,
+ pack( $StP{$engine->byte_size}, $new_index->offset ),
+ );
+ }
+
+ $sector->free;
+
+ $sector = $blist_cache{ ord( substr( $args->{key_md5}, $i, 1 ) ) };
+ $sector->find_md5( $args->{key_md5} );
+ }
+
+ return $sector;
+}
+
+sub get_class_offset {
+ my $self = shift;
+
+ my $e = $self->engine;
+ return unpack(
+ $StP{$e->byte_size},
+ $e->storage->read_at(
+ $self->offset + $self->base_size + 1 * $e->byte_size, $e->byte_size,
+ ),
);
+}
+
+sub get_classname {
+ my $self = shift;
+
+ my $class_offset = $self->get_class_offset;
+
+ return unless $class_offset;
+
+ return $self->engine->_load_sector( $class_offset )->data;
+}
+
+#XXX Add singleton handling here
+sub data {
+ my $self = shift;
+
+ my $new_obj = DBM::Deep->new({
+ type => $self->type,
+ base_offset => $self->offset,
+ staleness => $self->staleness,
+ storage => $self->engine->storage,
+ engine => $self->engine,
+ });
+
+ if ( $self->engine->storage->{autobless} ) {
+ my $classname = $self->get_classname;
+ if ( defined $classname ) {
+ bless $new_obj, $classname;
+ }
+ }
+
+ return $new_obj;
+}
+
+package DBM::Deep::Engine::Sector::BucketList;
+
+our @ISA = qw( DBM::Deep::Engine::Sector );
+
+sub _init {
+ my $self = shift;
+
+ my $engine = $self->engine;
+
+ unless ( $self->offset ) {
+ my $leftover = $self->size - $self->base_size;
+
+ $self->{offset} = $engine->_request_blist_sector( $self->size );
+ $engine->storage->print_at( $self->offset, $engine->SIG_BLIST ); # Sector type
+ # Skip staleness counter
+ $engine->storage->print_at( $self->offset + $self->base_size,
+ chr(0) x $leftover, # Zero-fill the data
+ );
+ }
+
+ if ( $self->{key_md5} ) {
+ $self->find_md5;
+ }
+
+ return $self;
+}
+
+sub size {
+ my $self = shift;
+ unless ( $self->{size} ) {
+ my $e = $self->engine;
+ # Base + numbuckets * bucketsize
+ $self->{size} = $self->base_size + $e->max_buckets * $self->bucket_size;
+ }
+ return $self->{size};
+}
+
+sub free_meth { return '_add_free_blist_sector' }
+
+sub bucket_size {
+ my $self = shift;
+ unless ( $self->{bucket_size} ) {
+ my $e = $self->engine;
+ # Key + head (location) + transactions (location + staleness-counter)
+ my $location_size = $e->byte_size + $e->byte_size + ($e->num_txns - 1) * ($e->byte_size + $STALE_SIZE);
+ $self->{bucket_size} = $e->hash_size + $location_size;
+ }
+ return $self->{bucket_size};
+}
+
+# XXX This is such a poor hack. I need to rethink this code.
+sub chopped_up {
+ my $self = shift;
+
+ my $e = $self->engine;
+
+ my @buckets;
+ foreach my $idx ( 0 .. $e->max_buckets - 1 ) {
+ my $spot = $self->offset + $self->base_size + $idx * $self->bucket_size;
+ my $md5 = $e->storage->read_at( $spot, $e->hash_size );
+
+ #XXX If we're chopping, why would we ever have the blank_md5?
+ last if $md5 eq $e->blank_md5;
+
+ my $rest = $e->storage->read_at( undef, $self->bucket_size - $e->hash_size );
+ push @buckets, [ $spot, $md5 . $rest ];
+ }
+
+ return @buckets;
+}
+
+sub write_at_next_open {
+ my $self = shift;
+ my ($entry) = @_;
+
+ #XXX This is such a hack!
+ $self->{_next_open} = 0 unless exists $self->{_next_open};
+
+ my $spot = $self->offset + $self->base_size + $self->{_next_open}++ * $self->bucket_size;
+ $self->engine->storage->print_at( $spot, $entry );
+
+ return $spot;
+}
+
+sub has_md5 {
+ my $self = shift;
+ unless ( exists $self->{found} ) {
+ $self->find_md5;
+ }
+ return $self->{found};
+}
+
+sub find_md5 {
+ my $self = shift;
+
+ $self->{found} = undef;
+ $self->{idx} = -1;
+
+ if ( @_ ) {
+ $self->{key_md5} = shift;
+ }
+
+ # If we don't have an MD5, then what are we supposed to do?
+ unless ( exists $self->{key_md5} ) {
+ DBM::Deep->_throw_error( "Cannot find_md5 without a key_md5 set" );
+ }
+
+ my $e = $self->engine;
+ foreach my $idx ( 0 .. $e->max_buckets - 1 ) {
+ my $potential = $e->storage->read_at(
+ $self->offset + $self->base_size + $idx * $self->bucket_size, $e->hash_size,
+ );
+
+ if ( $potential eq $e->blank_md5 ) {
+ $self->{idx} = $idx;
+ return;
+ }
+
+ if ( $potential eq $self->{key_md5} ) {
+ $self->{found} = 1;
+ $self->{idx} = $idx;
+ return;
+ }
+ }
return;
}
-sub _throw_error {
- die "DBM::Deep: $_[1]\n";
+sub write_md5 {
+ my $self = shift;
+ my ($args) = @_;
+
+ DBM::Deep->_throw_error( "write_md5: no key" ) unless exists $args->{key};
+ DBM::Deep->_throw_error( "write_md5: no key_md5" ) unless exists $args->{key_md5};
+ DBM::Deep->_throw_error( "write_md5: no value" ) unless exists $args->{value};
+
+ my $engine = $self->engine;
+
+ $args->{trans_id} = $engine->trans_id unless exists $args->{trans_id};
+
+ my $spot = $self->offset + $self->base_size + $self->{idx} * $self->bucket_size;
+ $engine->add_entry( $args->{trans_id}, $spot );
+
+ unless ($self->{found}) {
+ my $key_sector = DBM::Deep::Engine::Sector::Scalar->new({
+ engine => $engine,
+ data => $args->{key},
+ });
+
+ $engine->storage->print_at( $spot,
+ $args->{key_md5},
+ pack( $StP{$engine->byte_size}, $key_sector->offset ),
+ );
+ }
+
+ my $loc = $spot
+ + $engine->hash_size
+ + $engine->byte_size;
+
+ if ( $args->{trans_id} ) {
+ $loc += $engine->byte_size + ($args->{trans_id} - 1) * ( $engine->byte_size + $STALE_SIZE );
+
+ $engine->storage->print_at( $loc,
+ pack( $StP{$engine->byte_size}, $args->{value}->offset ),
+ pack( $StP{$STALE_SIZE}, $engine->get_txn_staleness_counter( $args->{trans_id} ) ),
+ );
+ }
+ else {
+ $engine->storage->print_at( $loc,
+ pack( $StP{$engine->byte_size}, $args->{value}->offset ),
+ );
+ }
+}
+
+sub mark_deleted {
+ my $self = shift;
+ my ($args) = @_;
+ $args ||= {};
+
+ my $engine = $self->engine;
+
+ $args->{trans_id} = $engine->trans_id unless exists $args->{trans_id};
+
+ my $spot = $self->offset + $self->base_size + $self->{idx} * $self->bucket_size;
+ $engine->add_entry( $args->{trans_id}, $spot );
+
+ my $loc = $spot
+ + $engine->hash_size
+ + $engine->byte_size;
+
+ if ( $args->{trans_id} ) {
+ $loc += $engine->byte_size + ($args->{trans_id} - 1) * ( $engine->byte_size + $STALE_SIZE );
+
+ $engine->storage->print_at( $loc,
+ pack( $StP{$engine->byte_size}, 1 ), # 1 is the marker for deleted
+ pack( $StP{$STALE_SIZE}, $engine->get_txn_staleness_counter( $args->{trans_id} ) ),
+ );
+ }
+ else {
+ $engine->storage->print_at( $loc,
+ pack( $StP{$engine->byte_size}, 1 ), # 1 is the marker for deleted
+ );
+ }
+
+}
+
+sub delete_md5 {
+ my $self = shift;
+ my ($args) = @_;
+
+ my $engine = $self->engine;
+ return undef unless $self->{found};
+
+ # Save the location so that we can free the data
+ my $location = $self->get_data_location_for({
+ allow_head => 0,
+ });
+ my $key_sector = $self->get_key_for;
+
+ my $spot = $self->offset + $self->base_size + $self->{idx} * $self->bucket_size;
+ $engine->storage->print_at( $spot,
+ $engine->storage->read_at(
+ $spot + $self->bucket_size,
+ $self->bucket_size * ( $engine->max_buckets - $self->{idx} - 1 ),
+ ),
+ chr(0) x $self->bucket_size,
+ );
+
+ $key_sector->free;
+
+ my $data_sector = $self->engine->_load_sector( $location );
+ my $data = $data_sector->data;
+ $data_sector->free;
+
+ return $data;
+}
+
+sub get_data_location_for {
+ my $self = shift;
+ my ($args) = @_;
+ $args ||= {};
+
+ $args->{allow_head} = 0 unless exists $args->{allow_head};
+ $args->{trans_id} = $self->engine->trans_id unless exists $args->{trans_id};
+ $args->{idx} = $self->{idx} unless exists $args->{idx};
+
+ my $e = $self->engine;
+
+ my $spot = $self->offset + $self->base_size
+ + $args->{idx} * $self->bucket_size
+ + $e->hash_size
+ + $e->byte_size;
+
+ if ( $args->{trans_id} ) {
+ $spot += $e->byte_size + ($args->{trans_id} - 1) * ( $e->byte_size + $STALE_SIZE );
+ }
+
+ my $buffer = $e->storage->read_at(
+ $spot,
+ $e->byte_size + $STALE_SIZE,
+ );
+ my ($loc, $staleness) = unpack( $StP{$e->byte_size} . ' ' . $StP{$STALE_SIZE}, $buffer );
+
+ if ( $args->{trans_id} ) {
+ # We have found an entry that is old, so get rid of it
+ if ( $staleness != (my $s = $e->get_txn_staleness_counter( $args->{trans_id} ) ) ) {
+ $e->storage->print_at(
+ $spot,
+ pack( $StP{$e->byte_size} . ' ' . $StP{$STALE_SIZE}, (0) x 2 ),
+ );
+ $loc = 0;
+ }
+ }
+
+ # If we're in a transaction and we never wrote to this location, try the
+ # HEAD instead.
+ if ( $args->{trans_id} && !$loc && $args->{allow_head} ) {
+ return $self->get_data_location_for({
+ trans_id => 0,
+ allow_head => 1,
+ idx => $args->{idx},
+ });
+ }
+ return $loc <= 1 ? 0 : $loc;
+}
+
+sub get_data_for {
+ my $self = shift;
+ my ($args) = @_;
+ $args ||= {};
+
+ return unless $self->{found};
+ my $location = $self->get_data_location_for({
+ allow_head => $args->{allow_head},
+ });
+ return $self->engine->_load_sector( $location );
+}
+
+sub get_key_for {
+ my $self = shift;
+ my ($idx) = @_;
+ $idx = $self->{idx} unless defined $idx;
+
+ if ( $idx >= $self->engine->max_buckets ) {
+ DBM::Deep->_throw_error( "get_key_for(): Attempting to retrieve $idx" );
+ }
+
+ my $location = $self->engine->storage->read_at(
+ $self->offset + $self->base_size + $idx * $self->bucket_size + $self->engine->hash_size,
+ $self->engine->byte_size,
+ );
+ $location = unpack( $StP{$self->engine->byte_size}, $location );
+ DBM::Deep->_throw_error( "get_key_for: No location?" ) unless $location;
+
+ return $self->engine->_load_sector( $location );
+}
+
+package DBM::Deep::Engine::Sector::Index;
+
+our @ISA = qw( DBM::Deep::Engine::Sector );
+
+sub _init {
+ my $self = shift;
+
+ my $engine = $self->engine;
+
+ unless ( $self->offset ) {
+ my $leftover = $self->size - $self->base_size;
+
+ $self->{offset} = $engine->_request_index_sector( $self->size );
+ $engine->storage->print_at( $self->offset, $engine->SIG_INDEX ); # Sector type
+ # Skip staleness counter
+ $engine->storage->print_at( $self->offset + $self->base_size,
+ chr(0) x $leftover, # Zero-fill the rest
+ );
+ }
+
+ return $self;
+}
+
+#XXX Change here
+sub size {
+ my $self = shift;
+ unless ( $self->{size} ) {
+ my $e = $self->engine;
+ $self->{size} = $self->base_size + $e->byte_size * $e->hash_chars;
+ }
+ return $self->{size};
+}
+
+sub free_meth { return '_add_free_index_sector' }
+
+sub free {
+ my $self = shift;
+ my $e = $self->engine;
+
+ for my $i ( 0 .. $e->hash_chars - 1 ) {
+ my $l = $self->get_entry( $i ) or next;
+ $e->_load_sector( $l )->free;
+ }
+
+ $self->SUPER::free();
+}
+
+sub _loc_for {
+ my $self = shift;
+ my ($idx) = @_;
+ return $self->offset + $self->base_size + $idx * $self->engine->byte_size;
+}
+
+sub get_entry {
+ my $self = shift;
+ my ($idx) = @_;
+
+ my $e = $self->engine;
+
+ DBM::Deep->_throw_error( "get_entry: Out of range ($idx)" )
+ if $idx < 0 || $idx >= $e->hash_chars;
+
+ return unpack(
+ $StP{$e->byte_size},
+ $e->storage->read_at( $self->_loc_for( $idx ), $e->byte_size ),
+ );
+}
+
+sub set_entry {
+ my $self = shift;
+ my ($idx, $loc) = @_;
+
+ my $e = $self->engine;
+
+ DBM::Deep->_throw_error( "set_entry: Out of range ($idx)" )
+ if $idx < 0 || $idx >= $e->hash_chars;
+
+ $self->engine->storage->print_at(
+ $self->_loc_for( $idx ),
+ pack( $StP{$e->byte_size}, $loc ),
+ );
}
1;
projects / dbsrgits/DBM-Deep.git / blobdiff