Commit | Line | Data |
b5aed31e |
1 | |
2 | package Tie::File; |
6fc0ea7e |
3 | require 5.005; |
6ae23f41 |
4 | use Carp ':DEFAULT', 'confess'; |
b5aed31e |
5 | use POSIX 'SEEK_SET'; |
6ae23f41 |
6 | use Fcntl 'O_CREAT', 'O_RDWR', 'LOCK_EX', 'LOCK_SH', 'O_WRONLY', 'O_RDONLY'; |
95f36366 |
7 | sub O_ACCMODE () { O_RDONLY | O_RDWR | O_WRONLY } |
b5aed31e |
8 | |
6ae23f41 |
9 | |
bc2f65e0 |
10 | $VERSION = "0.97"; |
b3fe5a4c |
11 | my $DEFAULT_MEMORY_SIZE = 1<<21; # 2 megabytes |
6fc0ea7e |
12 | my $DEFAULT_AUTODEFER_THRESHHOLD = 3; # 3 records |
13 | my $DEFAULT_AUTODEFER_FILELEN_THRESHHOLD = 65536; # 16 disk blocksful |
b3fe5a4c |
14 | |
6ae23f41 |
15 | my %good_opt = map {$_ => 1, "-$_" => 1} |
16 | qw(memory dw_size mode recsep discipline |
bc2f65e0 |
17 | autodefer autochomp autodefer_threshhold concurrent); |
b5aed31e |
18 | |
19 | sub TIEARRAY { |
20 | if (@_ % 2 != 0) { |
21 | croak "usage: tie \@array, $_[0], filename, [option => value]..."; |
22 | } |
23 | my ($pack, $file, %opts) = @_; |
24 | |
25 | # transform '-foo' keys into 'foo' keys |
26 | for my $key (keys %opts) { |
b3fe5a4c |
27 | unless ($good_opt{$key}) { |
28 | croak("$pack: Unrecognized option '$key'\n"); |
29 | } |
b5aed31e |
30 | my $okey = $key; |
31 | if ($key =~ s/^-+//) { |
32 | $opts{$key} = delete $opts{$okey}; |
33 | } |
34 | } |
35 | |
bc2f65e0 |
36 | if ($opts{concurrent}) { |
37 | croak("$pack: concurrent access not supported yet\n"); |
38 | } |
39 | |
b3fe5a4c |
40 | unless (defined $opts{memory}) { |
41 | # default is the larger of the default cache size and the |
42 | # deferred-write buffer size (if specified) |
43 | $opts{memory} = $DEFAULT_MEMORY_SIZE; |
6ae23f41 |
44 | $opts{memory} = $opts{dw_size} |
b3fe5a4c |
45 | if defined $opts{dw_size} && $opts{dw_size} > $DEFAULT_MEMORY_SIZE; |
57c7bc08 |
46 | # Dora Winifred Read |
b3fe5a4c |
47 | } |
48 | $opts{dw_size} = $opts{memory} unless defined $opts{dw_size}; |
49 | if ($opts{dw_size} > $opts{memory}) { |
50 | croak("$pack: dw_size may not be larger than total memory allocation\n"); |
51 | } |
57c7bc08 |
52 | # are we in deferred-write mode? |
53 | $opts{defer} = 0 unless defined $opts{defer}; |
54 | $opts{deferred} = {}; # no records are presently deferred |
b3fe5a4c |
55 | $opts{deferred_s} = 0; # count of total bytes in ->{deferred} |
6fc0ea7e |
56 | $opts{deferred_max} = -1; # empty |
b5aed31e |
57 | |
6ae23f41 |
58 | # What's a good way to arrange that this class can be overridden? |
59 | $opts{cache} = Tie::File::Cache->new($opts{memory}); |
6fc0ea7e |
60 | |
61 | # autodeferment is enabled by default |
62 | $opts{autodefer} = 1 unless defined $opts{autodefer}; |
63 | $opts{autodeferring} = 0; # but is not initially active |
64 | $opts{ad_history} = []; |
65 | $opts{autodefer_threshhold} = $DEFAULT_AUTODEFER_THRESHHOLD |
66 | unless defined $opts{autodefer_threshhold}; |
67 | $opts{autodefer_filelen_threshhold} = $DEFAULT_AUTODEFER_FILELEN_THRESHHOLD |
68 | unless defined $opts{autodefer_filelen_threshhold}; |
b5aed31e |
69 | |
70 | $opts{offsets} = [0]; |
71 | $opts{filename} = $file; |
b3fe5a4c |
72 | unless (defined $opts{recsep}) { |
73 | $opts{recsep} = _default_recsep(); |
74 | } |
b5aed31e |
75 | $opts{recseplen} = length($opts{recsep}); |
76 | if ($opts{recseplen} == 0) { |
77 | croak "Empty record separator not supported by $pack"; |
78 | } |
79 | |
0b28bc9a |
80 | $opts{autochomp} = 1 unless defined $opts{autochomp}; |
81 | |
27531ffb |
82 | $opts{mode} = O_CREAT|O_RDWR unless defined $opts{mode}; |
83 | $opts{rdonly} = (($opts{mode} & O_ACCMODE) == O_RDONLY); |
bf919750 |
84 | $opts{sawlastrec} = undef; |
27531ffb |
85 | |
fa408a35 |
86 | my $fh; |
b5aed31e |
87 | |
fa408a35 |
88 | if (UNIVERSAL::isa($file, 'GLOB')) { |
57c7bc08 |
89 | # We use 1 here on the theory that some systems |
90 | # may not indicate failure if we use 0. |
91 | # MSWin32 does not indicate failure with 0, but I don't know if |
92 | # it will indicate failure with 1 or not. |
93 | unless (seek $file, 1, SEEK_SET) { |
fa408a35 |
94 | croak "$pack: your filehandle does not appear to be seekable"; |
95 | } |
57c7bc08 |
96 | seek $file, 0, SEEK_SET # put it back |
97 | $fh = $file; # setting binmode is the user's problem |
fa408a35 |
98 | } elsif (ref $file) { |
99 | croak "usage: tie \@array, $pack, filename, [option => value]..."; |
100 | } else { |
f7363f9d |
101 | # $fh = \do { local *FH }; # XXX this is buggy |
e0cf91d5 |
102 | if ($] < 5.006) { |
103 | # perl 5.005 and earlier don't autovivify filehandles |
104 | require Symbol; |
105 | $fh = Symbol::gensym(); |
106 | } |
27531ffb |
107 | sysopen $fh, $file, $opts{mode}, 0666 or return; |
fa408a35 |
108 | binmode $fh; |
2a00e2a4 |
109 | ++$opts{ourfh}; |
fa408a35 |
110 | } |
b5aed31e |
111 | { my $ofh = select $fh; $| = 1; select $ofh } # autoflush on write |
b3fe5a4c |
112 | if (defined $opts{discipline} && $] >= 5.006) { |
113 | # This avoids a compile-time warning under 5.005 |
114 | eval 'binmode($fh, $opts{discipline})'; |
115 | croak $@ if $@ =~ /unknown discipline/i; |
116 | die if $@; |
117 | } |
b5aed31e |
118 | $opts{fh} = $fh; |
119 | |
120 | bless \%opts => $pack; |
121 | } |
122 | |
123 | sub FETCH { |
124 | my ($self, $n) = @_; |
6fc0ea7e |
125 | my $rec; |
126 | |
127 | # check the defer buffer |
6ae23f41 |
128 | $rec = $self->{deferred}{$n} if exists $self->{deferred}{$n}; |
129 | $rec = $self->_fetch($n) unless defined $rec; |
6fc0ea7e |
130 | |
6ae23f41 |
131 | # inlined _chomp1 |
132 | substr($rec, - $self->{recseplen}) = "" |
133 | if defined $rec && $self->{autochomp}; |
134 | $rec; |
0b28bc9a |
135 | } |
136 | |
137 | # Chomp many records in-place; return nothing useful |
138 | sub _chomp { |
139 | my $self = shift; |
140 | return unless $self->{autochomp}; |
141 | if ($self->{autochomp}) { |
142 | for (@_) { |
143 | next unless defined; |
144 | substr($_, - $self->{recseplen}) = ""; |
145 | } |
146 | } |
147 | } |
148 | |
149 | # Chomp one record in-place; return modified record |
150 | sub _chomp1 { |
151 | my ($self, $rec) = @_; |
152 | return $rec unless $self->{autochomp}; |
153 | return unless defined $rec; |
154 | substr($rec, - $self->{recseplen}) = ""; |
155 | $rec; |
156 | } |
157 | |
158 | sub _fetch { |
159 | my ($self, $n) = @_; |
b5aed31e |
160 | |
161 | # check the record cache |
6fc0ea7e |
162 | { my $cached = $self->{cache}->lookup($n); |
b5aed31e |
163 | return $cached if defined $cached; |
164 | } |
165 | |
27531ffb |
166 | if ($#{$self->{offsets}} < $n) { |
6ae23f41 |
167 | return if $self->{eof}; # request for record beyond end of file |
b5aed31e |
168 | my $o = $self->_fill_offsets_to($n); |
169 | # If it's still undefined, there is no such record, so return 'undef' |
170 | return unless defined $o; |
171 | } |
172 | |
173 | my $fh = $self->{FH}; |
174 | $self->_seek($n); # we can do this now that offsets is populated |
175 | my $rec = $self->_read_record; |
b3fe5a4c |
176 | |
177 | # If we happen to have just read the first record, check to see if |
178 | # the length of the record matches what 'tell' says. If not, Tie::File |
179 | # won't work, and should drop dead. |
180 | # |
181 | # if ($n == 0 && defined($rec) && tell($self->{fh}) != length($rec)) { |
182 | # if (defined $self->{discipline}) { |
183 | # croak "I/O discipline $self->{discipline} not supported"; |
184 | # } else { |
185 | # croak "File encoding not supported"; |
186 | # } |
187 | # } |
188 | |
6fc0ea7e |
189 | $self->{cache}->insert($n, $rec) if defined $rec && not $self->{flushing}; |
b5aed31e |
190 | $rec; |
191 | } |
192 | |
193 | sub STORE { |
194 | my ($self, $n, $rec) = @_; |
6fc0ea7e |
195 | die "STORE called from _check_integrity!" if $DIAGNOSTIC; |
b5aed31e |
196 | |
197 | $self->_fixrecs($rec); |
198 | |
6fc0ea7e |
199 | if ($self->{autodefer}) { |
200 | $self->_annotate_ad_history($n); |
201 | } |
202 | |
203 | return $self->_store_deferred($n, $rec) if $self->_is_deferring; |
204 | |
b5aed31e |
205 | |
206 | # We need this to decide whether the new record will fit |
207 | # It incidentally populates the offsets table |
208 | # Note we have to do this before we alter the cache |
6fc0ea7e |
209 | # 20020324 Wait, but this DOES alter the cache. TODO BUG? |
0b28bc9a |
210 | my $oldrec = $self->_fetch($n); |
b5aed31e |
211 | |
b5aed31e |
212 | if (not defined $oldrec) { |
213 | # We're storing a record beyond the end of the file |
51efdd02 |
214 | $self->_extend_file_to($n+1); |
b5aed31e |
215 | $oldrec = $self->{recsep}; |
216 | } |
6ae23f41 |
217 | # return if $oldrec eq $rec; # don't bother |
b5aed31e |
218 | my $len_diff = length($rec) - length($oldrec); |
219 | |
b3fe5a4c |
220 | # length($oldrec) here is not consistent with text mode TODO XXX BUG |
6ae23f41 |
221 | $self->_mtwrite($rec, $self->{offsets}[$n], length($oldrec)); |
222 | $self->_oadjust([$n, 1, $rec]); |
223 | $self->{cache}->update($n, $rec); |
b5aed31e |
224 | } |
225 | |
b3fe5a4c |
226 | sub _store_deferred { |
227 | my ($self, $n, $rec) = @_; |
6fc0ea7e |
228 | $self->{cache}->remove($n); |
b3fe5a4c |
229 | my $old_deferred = $self->{deferred}{$n}; |
6fc0ea7e |
230 | |
231 | if (defined $self->{deferred_max} && $n > $self->{deferred_max}) { |
232 | $self->{deferred_max} = $n; |
233 | } |
b3fe5a4c |
234 | $self->{deferred}{$n} = $rec; |
6fc0ea7e |
235 | |
236 | my $len_diff = length($rec); |
237 | $len_diff -= length($old_deferred) if defined $old_deferred; |
238 | $self->{deferred_s} += $len_diff; |
239 | $self->{cache}->adj_limit(-$len_diff); |
b3fe5a4c |
240 | if ($self->{deferred_s} > $self->{dw_size}) { |
57c7bc08 |
241 | $self->_flush; |
242 | } elsif ($self->_cache_too_full) { |
b3fe5a4c |
243 | $self->_cache_flush; |
244 | } |
245 | } |
246 | |
57c7bc08 |
247 | # Remove a single record from the deferred-write buffer without writing it |
248 | # The record need not be present |
249 | sub _delete_deferred { |
250 | my ($self, $n) = @_; |
251 | my $rec = delete $self->{deferred}{$n}; |
252 | return unless defined $rec; |
6fc0ea7e |
253 | |
254 | if (defined $self->{deferred_max} |
255 | && $n == $self->{deferred_max}) { |
256 | undef $self->{deferred_max}; |
257 | } |
258 | |
57c7bc08 |
259 | $self->{deferred_s} -= length $rec; |
6fc0ea7e |
260 | $self->{cache}->adj_limit(length $rec); |
57c7bc08 |
261 | } |
262 | |
b5aed31e |
263 | sub FETCHSIZE { |
264 | my $self = shift; |
6ae23f41 |
265 | my $n = $self->{eof} ? $#{$self->{offsets}} : $self->_fill_offsets; |
266 | |
6fc0ea7e |
267 | my $top_deferred = $self->_defer_max; |
268 | $n = $top_deferred+1 if defined $top_deferred && $n < $top_deferred+1; |
b5aed31e |
269 | $n; |
270 | } |
271 | |
272 | sub STORESIZE { |
273 | my ($self, $len) = @_; |
6fc0ea7e |
274 | |
275 | if ($self->{autodefer}) { |
276 | $self->_annotate_ad_history('STORESIZE'); |
277 | } |
278 | |
b5aed31e |
279 | my $olen = $self->FETCHSIZE; |
280 | return if $len == $olen; # Woo-hoo! |
281 | |
282 | # file gets longer |
283 | if ($len > $olen) { |
6fc0ea7e |
284 | if ($self->_is_deferring) { |
57c7bc08 |
285 | for ($olen .. $len-1) { |
286 | $self->_store_deferred($_, $self->{recsep}); |
287 | } |
288 | } else { |
289 | $self->_extend_file_to($len); |
290 | } |
b5aed31e |
291 | return; |
292 | } |
293 | |
294 | # file gets shorter |
6fc0ea7e |
295 | if ($self->_is_deferring) { |
296 | # TODO maybe replace this with map-plus-assignment? |
57c7bc08 |
297 | for (grep $_ >= $len, keys %{$self->{deferred}}) { |
298 | $self->_delete_deferred($_); |
299 | } |
6fc0ea7e |
300 | $self->{deferred_max} = $len-1; |
57c7bc08 |
301 | } |
302 | |
b5aed31e |
303 | $self->_seek($len); |
304 | $self->_chop_file; |
836d9961 |
305 | $#{$self->{offsets}} = $len; |
b3fe5a4c |
306 | # $self->{offsets}[0] = 0; # in case we just chopped this |
6fc0ea7e |
307 | |
bf919750 |
308 | $self->{cache}->remove(grep $_ >= $len, $self->{cache}->ckeys); |
b5aed31e |
309 | } |
310 | |
6ae23f41 |
311 | ### OPTIMIZE ME |
312 | ### It should not be necessary to do FETCHSIZE |
313 | ### Just seek to the end of the file. |
51efdd02 |
314 | sub PUSH { |
315 | my $self = shift; |
316 | $self->SPLICE($self->FETCHSIZE, scalar(@_), @_); |
6ae23f41 |
317 | |
318 | # No need to return: |
319 | # $self->FETCHSIZE; # because av.c takes care of this for me |
51efdd02 |
320 | } |
321 | |
322 | sub POP { |
323 | my $self = shift; |
7b6b3db1 |
324 | my $size = $self->FETCHSIZE; |
325 | return if $size == 0; |
326 | # print STDERR "# POPPITY POP POP POP\n"; |
327 | scalar $self->SPLICE($size-1, 1); |
51efdd02 |
328 | } |
329 | |
330 | sub SHIFT { |
331 | my $self = shift; |
332 | scalar $self->SPLICE(0, 1); |
333 | } |
334 | |
335 | sub UNSHIFT { |
336 | my $self = shift; |
337 | $self->SPLICE(0, 0, @_); |
57c7bc08 |
338 | # $self->FETCHSIZE; # av.c takes care of this for me |
51efdd02 |
339 | } |
340 | |
341 | sub CLEAR { |
51efdd02 |
342 | my $self = shift; |
6fc0ea7e |
343 | |
344 | if ($self->{autodefer}) { |
345 | $self->_annotate_ad_history('CLEAR'); |
346 | } |
347 | |
51efdd02 |
348 | $self->_seekb(0); |
349 | $self->_chop_file; |
6fc0ea7e |
350 | $self->{cache}->set_limit($self->{memory}); |
351 | $self->{cache}->empty; |
51efdd02 |
352 | @{$self->{offsets}} = (0); |
57c7bc08 |
353 | %{$self->{deferred}}= (); |
354 | $self->{deferred_s} = 0; |
6fc0ea7e |
355 | $self->{deferred_max} = -1; |
51efdd02 |
356 | } |
357 | |
358 | sub EXTEND { |
359 | my ($self, $n) = @_; |
57c7bc08 |
360 | |
361 | # No need to pre-extend anything in this case |
6fc0ea7e |
362 | return if $self->_is_deferring; |
57c7bc08 |
363 | |
51efdd02 |
364 | $self->_fill_offsets_to($n); |
365 | $self->_extend_file_to($n); |
366 | } |
367 | |
368 | sub DELETE { |
369 | my ($self, $n) = @_; |
6fc0ea7e |
370 | |
371 | if ($self->{autodefer}) { |
372 | $self->_annotate_ad_history('DELETE'); |
373 | } |
374 | |
51efdd02 |
375 | my $lastrec = $self->FETCHSIZE-1; |
57c7bc08 |
376 | my $rec = $self->FETCH($n); |
6fc0ea7e |
377 | $self->_delete_deferred($n) if $self->_is_deferring; |
51efdd02 |
378 | if ($n == $lastrec) { |
379 | $self->_seek($n); |
380 | $self->_chop_file; |
fa408a35 |
381 | $#{$self->{offsets}}--; |
6fc0ea7e |
382 | $self->{cache}->remove($n); |
51efdd02 |
383 | # perhaps in this case I should also remove trailing null records? |
57c7bc08 |
384 | # 20020316 |
385 | # Note that delete @a[-3..-1] deletes the records in the wrong order, |
386 | # so we only chop the very last one out of the file. We could repair this |
387 | # by tracking deleted records inside the object. |
388 | } elsif ($n < $lastrec) { |
51efdd02 |
389 | $self->STORE($n, ""); |
390 | } |
57c7bc08 |
391 | $rec; |
51efdd02 |
392 | } |
393 | |
394 | sub EXISTS { |
395 | my ($self, $n) = @_; |
57c7bc08 |
396 | return 1 if exists $self->{deferred}{$n}; |
57c7bc08 |
397 | $n < $self->FETCHSIZE; |
51efdd02 |
398 | } |
399 | |
b5aed31e |
400 | sub SPLICE { |
b3fe5a4c |
401 | my $self = shift; |
6fc0ea7e |
402 | |
403 | if ($self->{autodefer}) { |
404 | $self->_annotate_ad_history('SPLICE'); |
405 | } |
406 | |
407 | $self->_flush if $self->_is_deferring; # move this up? |
0b28bc9a |
408 | if (wantarray) { |
409 | $self->_chomp(my @a = $self->_splice(@_)); |
410 | @a; |
411 | } else { |
412 | $self->_chomp1(scalar $self->_splice(@_)); |
413 | } |
b3fe5a4c |
414 | } |
415 | |
416 | sub DESTROY { |
57c7bc08 |
417 | my $self = shift; |
6fc0ea7e |
418 | $self->flush if $self->_is_deferring; |
419 | $self->{cache}->delink if defined $self->{cache}; # break circular link |
2a00e2a4 |
420 | if ($self->{fh} and $self->{ourfh}) { |
421 | delete $self->{ourfh}; |
422 | close delete $self->{fh}; |
2a00e2a4 |
423 | } |
b3fe5a4c |
424 | } |
425 | |
426 | sub _splice { |
b5aed31e |
427 | my ($self, $pos, $nrecs, @data) = @_; |
428 | my @result; |
429 | |
7b6b3db1 |
430 | $pos = 0 unless defined $pos; |
431 | |
432 | # Deal with negative and other out-of-range positions |
433 | # Also set default for $nrecs |
51efdd02 |
434 | { |
435 | my $oldsize = $self->FETCHSIZE; |
7b6b3db1 |
436 | $nrecs = $oldsize unless defined $nrecs; |
51efdd02 |
437 | my $oldpos = $pos; |
438 | |
439 | if ($pos < 0) { |
440 | $pos += $oldsize; |
441 | if ($pos < 0) { |
442 | croak "Modification of non-creatable array value attempted, subscript $oldpos"; |
443 | } |
444 | } |
445 | |
446 | if ($pos > $oldsize) { |
447 | return unless @data; |
448 | $pos = $oldsize; # This is what perl does for normal arrays |
449 | } |
bf919750 |
450 | |
451 | # The manual is very unclear here |
452 | if ($nrecs < 0) { |
453 | $nrecs = $oldsize - $pos + $nrecs; |
454 | $nrecs = 0 if $nrecs < 0; |
455 | } |
6ae23f41 |
456 | |
457 | # nrecs is too big---it really means "until the end" |
458 | # 20030507 |
459 | if ($nrecs + $pos > $oldsize) { |
460 | $nrecs = $oldsize - $pos; |
461 | } |
51efdd02 |
462 | } |
b5aed31e |
463 | |
464 | $self->_fixrecs(@data); |
465 | my $data = join '', @data; |
466 | my $datalen = length $data; |
467 | my $oldlen = 0; |
468 | |
469 | # compute length of data being removed |
470 | for ($pos .. $pos+$nrecs-1) { |
27531ffb |
471 | last unless defined $self->_fill_offsets_to($_); |
0b28bc9a |
472 | my $rec = $self->_fetch($_); |
b5aed31e |
473 | last unless defined $rec; |
474 | push @result, $rec; |
6fc0ea7e |
475 | |
476 | # Why don't we just use length($rec) here? |
477 | # Because that record might have come from the cache. _splice |
478 | # might have been called to flush out the deferred-write records, |
27531ffb |
479 | # and in this case length($rec) is the length of the record to be |
480 | # *written*, not the length of the actual record in the file. But |
481 | # the offsets are still true. 20020322 |
6fc0ea7e |
482 | $oldlen += $self->{offsets}[$_+1] - $self->{offsets}[$_] |
483 | if defined $self->{offsets}[$_+1]; |
b5aed31e |
484 | } |
6ae23f41 |
485 | $self->_fill_offsets_to($pos+$nrecs); |
b5aed31e |
486 | |
51efdd02 |
487 | # Modify the file |
6ae23f41 |
488 | $self->_mtwrite($data, $self->{offsets}[$pos], $oldlen); |
489 | # Adjust the offsets table |
490 | $self->_oadjust([$pos, $nrecs, @data]); |
491 | |
492 | { # Take this read cache stuff out into a separate function |
493 | # You made a half-attempt to put it into _oadjust. |
494 | # Finish something like that up eventually. |
495 | # STORE also needs to do something similarish |
496 | |
497 | # update the read cache, part 1 |
498 | # modified records |
499 | for ($pos .. $pos+$nrecs-1) { |
500 | my $new = $data[$_-$pos]; |
501 | if (defined $new) { |
502 | $self->{cache}->update($_, $new); |
503 | } else { |
504 | $self->{cache}->remove($_); |
505 | } |
b5aed31e |
506 | } |
6ae23f41 |
507 | |
508 | # update the read cache, part 2 |
509 | # moved records - records past the site of the change |
510 | # need to be renumbered |
511 | # Maybe merge this with the previous block? |
512 | { |
513 | my @oldkeys = grep $_ >= $pos + $nrecs, $self->{cache}->ckeys; |
514 | my @newkeys = map $_-$nrecs+@data, @oldkeys; |
515 | $self->{cache}->rekey(\@oldkeys, \@newkeys); |
b5aed31e |
516 | } |
6fc0ea7e |
517 | |
6ae23f41 |
518 | # Now there might be too much data in the cache, if we spliced out |
519 | # some short records and spliced in some long ones. If so, flush |
520 | # the cache. |
521 | $self->_cache_flush; |
b5aed31e |
522 | } |
b5aed31e |
523 | |
51efdd02 |
524 | # Yes, the return value of 'splice' *is* actually this complicated |
525 | wantarray ? @result : @result ? $result[-1] : undef; |
b5aed31e |
526 | } |
527 | |
6ae23f41 |
528 | |
b5aed31e |
529 | # write data into the file |
6ae23f41 |
530 | # $data is the data to be written. |
b5aed31e |
531 | # it should be written at position $pos, and should overwrite |
532 | # exactly $len of the following bytes. |
533 | # Note that if length($data) > $len, the subsequent bytes will have to |
534 | # be moved up, and if length($data) < $len, they will have to |
535 | # be moved down |
536 | sub _twrite { |
537 | my ($self, $data, $pos, $len) = @_; |
538 | |
539 | unless (defined $pos) { |
540 | die "\$pos was undefined in _twrite"; |
541 | } |
542 | |
543 | my $len_diff = length($data) - $len; |
544 | |
545 | if ($len_diff == 0) { # Woo-hoo! |
546 | my $fh = $self->{fh}; |
547 | $self->_seekb($pos); |
548 | $self->_write_record($data); |
549 | return; # well, that was easy. |
550 | } |
551 | |
552 | # the two records are of different lengths |
553 | # our strategy here: rewrite the tail of the file, |
554 | # reading ahead one buffer at a time |
555 | # $bufsize is required to be at least as large as the data we're overwriting |
556 | my $bufsize = _bufsize($len_diff); |
557 | my ($writepos, $readpos) = ($pos, $pos+$len); |
51efdd02 |
558 | my $next_block; |
6fc0ea7e |
559 | my $more_data; |
b5aed31e |
560 | |
561 | # Seems like there ought to be a way to avoid the repeated code |
562 | # and the special case here. The read(1) is also a little weird. |
563 | # Think about this. |
564 | do { |
565 | $self->_seekb($readpos); |
51efdd02 |
566 | my $br = read $self->{fh}, $next_block, $bufsize; |
6fc0ea7e |
567 | $more_data = read $self->{fh}, my($dummy), 1; |
b5aed31e |
568 | $self->_seekb($writepos); |
569 | $self->_write_record($data); |
570 | $readpos += $br; |
571 | $writepos += length $data; |
572 | $data = $next_block; |
6ae23f41 |
573 | } while $more_data; |
51efdd02 |
574 | $self->_seekb($writepos); |
575 | $self->_write_record($next_block); |
b5aed31e |
576 | |
577 | # There might be leftover data at the end of the file |
578 | $self->_chop_file if $len_diff < 0; |
579 | } |
580 | |
6ae23f41 |
581 | # _iwrite(D, S, E) |
582 | # Insert text D at position S. |
583 | # Let C = E-S-|D|. If C < 0; die. |
584 | # Data in [S,S+C) is copied to [S+D,S+D+C) = [S+D,E). |
585 | # Data in [S+C = E-D, E) is returned. Data in [E, oo) is untouched. |
586 | # |
587 | # In a later version, don't read the entire intervening area into |
588 | # memory at once; do the copying block by block. |
589 | sub _iwrite { |
590 | my $self = shift; |
591 | my ($D, $s, $e) = @_; |
592 | my $d = length $D; |
593 | my $c = $e-$s-$d; |
594 | local *FH = $self->{fh}; |
595 | confess "Not enough space to insert $d bytes between $s and $e" |
596 | if $c < 0; |
597 | confess "[$s,$e) is an invalid insertion range" if $e < $s; |
598 | |
599 | $self->_seekb($s); |
600 | read FH, my $buf, $e-$s; |
601 | |
602 | $D .= substr($buf, 0, $c, ""); |
603 | |
604 | $self->_seekb($s); |
605 | $self->_write_record($D); |
606 | |
607 | return $buf; |
608 | } |
609 | |
610 | # Like _twrite, but the data-pos-len triple may be repeated; you may |
611 | # write several chunks. All the writing will be done in |
612 | # one pass. Chunks SHALL be in ascending order and SHALL NOT overlap. |
613 | sub _mtwrite { |
614 | my $self = shift; |
615 | my $unwritten = ""; |
616 | my $delta = 0; |
617 | |
618 | @_ % 3 == 0 |
619 | or die "Arguments to _mtwrite did not come in groups of three"; |
620 | |
621 | while (@_) { |
622 | my ($data, $pos, $len) = splice @_, 0, 3; |
623 | my $end = $pos + $len; # The OLD end of the segment to be replaced |
624 | $data = $unwritten . $data; |
625 | $delta -= length($unwritten); |
626 | $unwritten = ""; |
627 | $pos += $delta; # This is where the data goes now |
628 | my $dlen = length $data; |
629 | $self->_seekb($pos); |
630 | if ($len >= $dlen) { # the data will fit |
631 | $self->_write_record($data); |
632 | $delta += ($dlen - $len); # everything following moves down by this much |
633 | $data = ""; # All the data in the buffer has been written |
634 | } else { # won't fit |
635 | my $writable = substr($data, 0, $len - $delta, ""); |
636 | $self->_write_record($writable); |
637 | $delta += ($dlen - $len); # everything following moves down by this much |
638 | } |
639 | |
640 | # At this point we've written some but maybe not all of the data. |
641 | # There might be a gap to close up, or $data might still contain a |
642 | # bunch of unwritten data that didn't fit. |
643 | my $ndlen = length $data; |
644 | if ($delta == 0) { |
645 | $self->_write_record($data); |
646 | } elsif ($delta < 0) { |
647 | # upcopy (close up gap) |
648 | if (@_) { |
649 | $self->_upcopy($end, $end + $delta, $_[1] - $end); |
650 | } else { |
651 | $self->_upcopy($end, $end + $delta); |
652 | } |
653 | } else { |
654 | # downcopy (insert data that didn't fit; replace this data in memory |
655 | # with _later_ data that doesn't fit) |
656 | if (@_) { |
657 | $unwritten = $self->_downcopy($data, $end, $_[1] - $end); |
658 | } else { |
659 | # Make the file longer to accomodate the last segment that doesn' |
660 | $unwritten = $self->_downcopy($data, $end); |
661 | } |
662 | } |
663 | } |
664 | } |
665 | |
666 | # Copy block of data of length $len from position $spos to position $dpos |
667 | # $dpos must be <= $spos |
668 | # |
669 | # If $len is undefined, go all the way to the end of the file |
670 | # and then truncate it ($spos - $dpos bytes will be removed) |
671 | sub _upcopy { |
672 | my $blocksize = 8192; |
673 | my ($self, $spos, $dpos, $len) = @_; |
674 | if ($dpos > $spos) { |
675 | die "source ($spos) was upstream of destination ($dpos) in _upcopy"; |
676 | } elsif ($dpos == $spos) { |
677 | return; |
678 | } |
679 | |
680 | while (! defined ($len) || $len > 0) { |
681 | my $readsize = ! defined($len) ? $blocksize |
682 | : $len > $blocksize ? $blocksize |
683 | : $len; |
684 | |
685 | my $fh = $self->{fh}; |
686 | $self->_seekb($spos); |
86e5a81e |
687 | my $bytes_read = read $fh, my($data), $readsize; |
6ae23f41 |
688 | $self->_seekb($dpos); |
689 | if ($data eq "") { |
690 | $self->_chop_file; |
691 | last; |
692 | } |
693 | $self->_write_record($data); |
694 | $spos += $bytes_read; |
695 | $dpos += $bytes_read; |
696 | $len -= $bytes_read if defined $len; |
697 | } |
698 | } |
699 | |
700 | # Write $data into a block of length $len at position $pos, |
701 | # moving everything in the block forwards to make room. |
702 | # Instead of writing the last length($data) bytes from the block |
703 | # (because there isn't room for them any longer) return them. |
bc2f65e0 |
704 | # |
705 | # Undefined $len means 'until the end of the file' |
6ae23f41 |
706 | sub _downcopy { |
707 | my $blocksize = 8192; |
708 | my ($self, $data, $pos, $len) = @_; |
709 | my $fh = $self->{fh}; |
710 | |
711 | while (! defined $len || $len > 0) { |
712 | my $readsize = ! defined($len) ? $blocksize |
713 | : $len > $blocksize? $blocksize : $len; |
714 | $self->_seekb($pos); |
86e5a81e |
715 | read $fh, my($old), $readsize; |
bc2f65e0 |
716 | my $last_read_was_short = length($old) < $readsize; |
6ae23f41 |
717 | $data .= $old; |
bc2f65e0 |
718 | my $writable; |
719 | if ($last_read_was_short) { |
720 | # If last read was short, then $data now contains the entire rest |
721 | # of the file, so there's no need to write only one block of it |
722 | $writable = $data; |
723 | $data = ""; |
724 | } else { |
725 | $writable = substr($data, 0, $readsize, ""); |
726 | } |
6ae23f41 |
727 | last if $writable eq ""; |
bc2f65e0 |
728 | $self->_seekb($pos); |
6ae23f41 |
729 | $self->_write_record($writable); |
730 | $len -= $readsize if defined $len; |
731 | $pos += $readsize; |
732 | } |
733 | return $data; |
734 | } |
735 | |
736 | # Adjust the object data structures following an '_mtwrite' |
737 | # Arguments are |
738 | # [$pos, $nrecs, @length] items |
739 | # indicating that $nrecs records were removed at $recpos (a record offset) |
740 | # and replaced with records of length @length... |
741 | # Arguments guarantee that $recpos is strictly increasing. |
742 | # No return value |
743 | sub _oadjust { |
744 | my $self = shift; |
745 | my $delta = 0; |
746 | my $delta_recs = 0; |
747 | my $prev_end = -1; |
748 | my %newkeys; |
749 | |
750 | for (@_) { |
751 | my ($pos, $nrecs, @data) = @$_; |
752 | $pos += $delta_recs; |
753 | |
754 | # Adjust the offsets of the records after the previous batch up |
755 | # to the first new one of this batch |
756 | for my $i ($prev_end+2 .. $pos - 1) { |
757 | $self->{offsets}[$i] += $delta; |
758 | $newkey{$i} = $i + $delta_recs; |
759 | } |
760 | |
761 | $prev_end = $pos + @data - 1; # last record moved on this pass |
762 | |
763 | # Remove the offsets for the removed records; |
764 | # replace with the offsets for the inserted records |
765 | my @newoff = ($self->{offsets}[$pos] + $delta); |
766 | for my $i (0 .. $#data) { |
767 | my $newlen = length $data[$i]; |
768 | push @newoff, $newoff[$i] + $newlen; |
769 | $delta += $newlen; |
770 | } |
771 | |
772 | for my $i ($pos .. $pos+$nrecs-1) { |
773 | last if $i+1 > $#{$self->{offsets}}; |
774 | my $oldlen = $self->{offsets}[$i+1] - $self->{offsets}[$i]; |
775 | $delta -= $oldlen; |
776 | } |
777 | |
778 | # # also this data has changed, so update it in the cache |
779 | # for (0 .. $#data) { |
780 | # $self->{cache}->update($pos + $_, $data[$_]); |
781 | # } |
782 | # if ($delta_recs) { |
783 | # my @oldkeys = grep $_ >= $pos + @data, $self->{cache}->ckeys; |
784 | # my @newkeys = map $_ + $delta_recs, @oldkeys; |
785 | # $self->{cache}->rekey(\@oldkeys, \@newkeys); |
786 | # } |
787 | |
788 | # replace old offsets with new |
789 | splice @{$self->{offsets}}, $pos, $nrecs+1, @newoff; |
790 | # What if we just spliced out the end of the offsets table? |
791 | # shouldn't we clear $self->{eof}? Test for this XXX BUG TODO |
792 | |
793 | $delta_recs += @data - $nrecs; # net change in total number of records |
794 | } |
795 | |
796 | # The trailing records at the very end of the file |
797 | if ($delta) { |
798 | for my $i ($prev_end+2 .. $#{$self->{offsets}}) { |
799 | $self->{offsets}[$i] += $delta; |
800 | } |
801 | } |
802 | |
803 | # If we scrubbed out all known offsets, regenerate the trivial table |
804 | # that knows that the file does indeed start at 0. |
805 | $self->{offsets}[0] = 0 unless @{$self->{offsets}}; |
806 | # If the file got longer, the offsets table is no longer complete |
807 | # $self->{eof} = 0 if $delta_recs > 0; |
808 | |
809 | # Now there might be too much data in the cache, if we spliced out |
810 | # some short records and spliced in some long ones. If so, flush |
811 | # the cache. |
812 | $self->_cache_flush; |
813 | } |
814 | |
b5aed31e |
815 | # If a record does not already end with the appropriate terminator |
816 | # string, append one. |
817 | sub _fixrecs { |
818 | my $self = shift; |
819 | for (@_) { |
27531ffb |
820 | $_ = "" unless defined $_; |
b5aed31e |
821 | $_ .= $self->{recsep} |
822 | unless substr($_, - $self->{recseplen}) eq $self->{recsep}; |
823 | } |
824 | } |
825 | |
57c7bc08 |
826 | |
827 | ################################################################ |
828 | # |
829 | # Basic read, write, and seek |
830 | # |
831 | |
b5aed31e |
832 | # seek to the beginning of record #$n |
833 | # Assumes that the offsets table is already correctly populated |
834 | # |
835 | # Note that $n=-1 has a special meaning here: It means the start of |
836 | # the last known record; this may or may not be the very last record |
837 | # in the file, depending on whether the offsets table is fully populated. |
838 | # |
839 | sub _seek { |
840 | my ($self, $n) = @_; |
841 | my $o = $self->{offsets}[$n]; |
842 | defined($o) |
843 | or confess("logic error: undefined offset for record $n"); |
844 | seek $self->{fh}, $o, SEEK_SET |
6ae23f41 |
845 | or confess "Couldn't seek filehandle: $!"; # "Should never happen." |
b5aed31e |
846 | } |
847 | |
6ae23f41 |
848 | # seek to byte $b in the file |
b5aed31e |
849 | sub _seekb { |
850 | my ($self, $b) = @_; |
851 | seek $self->{fh}, $b, SEEK_SET |
852 | or die "Couldn't seek filehandle: $!"; # "Should never happen." |
853 | } |
854 | |
855 | # populate the offsets table up to the beginning of record $n |
856 | # return the offset of record $n |
857 | sub _fill_offsets_to { |
858 | my ($self, $n) = @_; |
27531ffb |
859 | |
860 | return $self->{offsets}[$n] if $self->{eof}; |
861 | |
b5aed31e |
862 | my $fh = $self->{fh}; |
863 | local *OFF = $self->{offsets}; |
864 | my $rec; |
865 | |
866 | until ($#OFF >= $n) { |
b5aed31e |
867 | $self->_seek(-1); # tricky -- see comment at _seek |
868 | $rec = $self->_read_record; |
869 | if (defined $rec) { |
6ae23f41 |
870 | push @OFF, int(tell $fh); # Tels says that int() saves memory here |
b5aed31e |
871 | } else { |
27531ffb |
872 | $self->{eof} = 1; |
b5aed31e |
873 | return; # It turns out there is no such record |
874 | } |
875 | } |
876 | |
877 | # we have now read all the records up to record n-1, |
878 | # so we can return the offset of record n |
6ae23f41 |
879 | $OFF[$n]; |
880 | } |
881 | |
882 | sub _fill_offsets { |
883 | my ($self) = @_; |
884 | |
885 | my $fh = $self->{fh}; |
886 | local *OFF = $self->{offsets}; |
887 | |
888 | $self->_seek(-1); # tricky -- see comment at _seek |
889 | |
890 | # Tels says that inlining read_record() would make this loop |
891 | # five times faster. 20030508 |
892 | while ( defined $self->_read_record()) { |
893 | # int() saves us memory here |
894 | push @OFF, int(tell $fh); |
895 | } |
896 | |
897 | $self->{eof} = 1; |
898 | $#OFF; |
b5aed31e |
899 | } |
900 | |
901 | # assumes that $rec is already suitably terminated |
902 | sub _write_record { |
903 | my ($self, $rec) = @_; |
904 | my $fh = $self->{fh}; |
bf919750 |
905 | local $\ = ""; |
b5aed31e |
906 | print $fh $rec |
907 | or die "Couldn't write record: $!"; # "Should never happen." |
27531ffb |
908 | # $self->{_written} += length($rec); |
b5aed31e |
909 | } |
910 | |
911 | sub _read_record { |
912 | my $self = shift; |
913 | my $rec; |
914 | { local $/ = $self->{recsep}; |
915 | my $fh = $self->{fh}; |
916 | $rec = <$fh>; |
917 | } |
27531ffb |
918 | return unless defined $rec; |
29aee836 |
919 | if (substr($rec, -$self->{recseplen}) ne $self->{recsep}) { |
27531ffb |
920 | # improperly terminated final record --- quietly fix it. |
921 | # my $ac = substr($rec, -$self->{recseplen}); |
922 | # $ac =~ s/\n/\\n/g; |
bf919750 |
923 | $self->{sawlastrec} = 1; |
27531ffb |
924 | unless ($self->{rdonly}) { |
bf919750 |
925 | local $\ = ""; |
27531ffb |
926 | my $fh = $self->{fh}; |
927 | print $fh $self->{recsep}; |
928 | } |
929 | $rec .= $self->{recsep}; |
930 | } |
931 | # $self->{_read} += length($rec) if defined $rec; |
b5aed31e |
932 | $rec; |
933 | } |
934 | |
6fc0ea7e |
935 | sub _rw_stats { |
27531ffb |
936 | my $self = shift; |
6fc0ea7e |
937 | @{$self}{'_read', '_written'}; |
938 | } |
939 | |
57c7bc08 |
940 | ################################################################ |
941 | # |
942 | # Read cache management |
943 | |
6fc0ea7e |
944 | sub _cache_flush { |
945 | my ($self) = @_; |
946 | $self->{cache}->reduce_size_to($self->{memory} - $self->{deferred_s}); |
b5aed31e |
947 | } |
948 | |
57c7bc08 |
949 | sub _cache_too_full { |
950 | my $self = shift; |
6fc0ea7e |
951 | $self->{cache}->bytes + $self->{deferred_s} >= $self->{memory}; |
b5aed31e |
952 | } |
953 | |
57c7bc08 |
954 | ################################################################ |
955 | # |
956 | # File custodial services |
957 | # |
958 | |
959 | |
b5aed31e |
960 | # We have read to the end of the file and have the offsets table |
961 | # entirely populated. Now we need to write a new record beyond |
962 | # the end of the file. We prepare for this by writing |
963 | # empty records into the file up to the position we want |
51efdd02 |
964 | # |
965 | # assumes that the offsets table already contains the offset of record $n, |
966 | # if it exists, and extends to the end of the file if not. |
b5aed31e |
967 | sub _extend_file_to { |
968 | my ($self, $n) = @_; |
969 | $self->_seek(-1); # position after the end of the last record |
970 | my $pos = $self->{offsets}[-1]; |
971 | |
972 | # the offsets table has one entry more than the total number of records |
6fc0ea7e |
973 | my $extras = $n - $#{$self->{offsets}}; |
b5aed31e |
974 | |
975 | # Todo : just use $self->{recsep} x $extras here? |
976 | while ($extras-- > 0) { |
977 | $self->_write_record($self->{recsep}); |
6ae23f41 |
978 | push @{$self->{offsets}}, int(tell $self->{fh}); |
b5aed31e |
979 | } |
980 | } |
981 | |
982 | # Truncate the file at the current position |
983 | sub _chop_file { |
984 | my $self = shift; |
985 | truncate $self->{fh}, tell($self->{fh}); |
986 | } |
987 | |
57c7bc08 |
988 | |
b5aed31e |
989 | # compute the size of a buffer suitable for moving |
990 | # all the data in a file forward $n bytes |
991 | # ($n may be negative) |
992 | # The result should be at least $n. |
993 | sub _bufsize { |
994 | my $n = shift; |
6ae23f41 |
995 | return 8192 if $n <= 0; |
b5aed31e |
996 | my $b = $n & ~8191; |
997 | $b += 8192 if $n & 8191; |
998 | $b; |
999 | } |
1000 | |
57c7bc08 |
1001 | ################################################################ |
1002 | # |
1003 | # Miscellaneous public methods |
1004 | # |
1005 | |
51efdd02 |
1006 | # Lock the file |
1007 | sub flock { |
1008 | my ($self, $op) = @_; |
1009 | unless (@_ <= 3) { |
1010 | my $pack = ref $self; |
1011 | croak "Usage: $pack\->flock([OPERATION])"; |
1012 | } |
1013 | my $fh = $self->{fh}; |
1014 | $op = LOCK_EX unless defined $op; |
6ae23f41 |
1015 | my $locked = flock $fh, $op; |
1016 | |
1017 | if ($locked && ($op & (LOCK_EX | LOCK_SH))) { |
1018 | # If you're locking the file, then presumably it's because |
1019 | # there might have been a write access by another process. |
1020 | # In that case, the read cache contents and the offsets table |
1021 | # might be invalid, so discard them. 20030508 |
1022 | $self->{offsets} = [0]; |
1023 | $self->{cache}->empty; |
1024 | } |
1025 | |
1026 | $locked; |
51efdd02 |
1027 | } |
b5aed31e |
1028 | |
0b28bc9a |
1029 | # Get/set autochomp option |
1030 | sub autochomp { |
1031 | my $self = shift; |
1032 | if (@_) { |
1033 | my $old = $self->{autochomp}; |
1034 | $self->{autochomp} = shift; |
1035 | $old; |
1036 | } else { |
1037 | $self->{autochomp}; |
1038 | } |
1039 | } |
1040 | |
6ae23f41 |
1041 | # Get offset table entries; returns offset of nth record |
1042 | sub offset { |
1043 | my ($self, $n) = @_; |
1044 | |
1045 | if ($#{$self->{offsets}} < $n) { |
1046 | return if $self->{eof}; # request for record beyond the end of file |
1047 | my $o = $self->_fill_offsets_to($n); |
1048 | # If it's still undefined, there is no such record, so return 'undef' |
1049 | return unless defined $o; |
1050 | } |
1051 | |
1052 | $self->{offsets}[$n]; |
1053 | } |
1054 | |
1055 | sub discard_offsets { |
1056 | my $self = shift; |
1057 | $self->{offsets} = [0]; |
1058 | } |
1059 | |
57c7bc08 |
1060 | ################################################################ |
1061 | # |
1062 | # Matters related to deferred writing |
1063 | # |
1064 | |
1065 | # Defer writes |
1066 | sub defer { |
1067 | my $self = shift; |
6fc0ea7e |
1068 | $self->_stop_autodeferring; |
1069 | @{$self->{ad_history}} = (); |
57c7bc08 |
1070 | $self->{defer} = 1; |
1071 | } |
1072 | |
b3fe5a4c |
1073 | # Flush deferred writes |
1074 | # |
1075 | # This could be better optimized to write the file in one pass, instead |
1076 | # of one pass per block of records. But that will require modifications |
6ae23f41 |
1077 | # to _twrite, so I should have a good _twrite test suite first. |
b3fe5a4c |
1078 | sub flush { |
1079 | my $self = shift; |
1080 | |
1081 | $self->_flush; |
1082 | $self->{defer} = 0; |
1083 | } |
1084 | |
6ae23f41 |
1085 | sub _old_flush { |
b3fe5a4c |
1086 | my $self = shift; |
1087 | my @writable = sort {$a<=>$b} (keys %{$self->{deferred}}); |
6ae23f41 |
1088 | |
b3fe5a4c |
1089 | while (@writable) { |
1090 | # gather all consecutive records from the front of @writable |
1091 | my $first_rec = shift @writable; |
1092 | my $last_rec = $first_rec+1; |
1093 | ++$last_rec, shift @writable while @writable && $last_rec == $writable[0]; |
1094 | --$last_rec; |
1095 | $self->_fill_offsets_to($last_rec); |
1096 | $self->_extend_file_to($last_rec); |
1097 | $self->_splice($first_rec, $last_rec-$first_rec+1, |
1098 | @{$self->{deferred}}{$first_rec .. $last_rec}); |
1099 | } |
1100 | |
57c7bc08 |
1101 | $self->_discard; # clear out defered-write-cache |
b3fe5a4c |
1102 | } |
1103 | |
6ae23f41 |
1104 | sub _flush { |
1105 | my $self = shift; |
1106 | my @writable = sort {$a<=>$b} (keys %{$self->{deferred}}); |
1107 | my @args; |
1108 | my @adjust; |
1109 | |
1110 | while (@writable) { |
1111 | # gather all consecutive records from the front of @writable |
1112 | my $first_rec = shift @writable; |
1113 | my $last_rec = $first_rec+1; |
1114 | ++$last_rec, shift @writable while @writable && $last_rec == $writable[0]; |
1115 | --$last_rec; |
1116 | my $end = $self->_fill_offsets_to($last_rec+1); |
1117 | if (not defined $end) { |
1118 | $self->_extend_file_to($last_rec); |
1119 | $end = $self->{offsets}[$last_rec]; |
1120 | } |
1121 | my ($start) = $self->{offsets}[$first_rec]; |
1122 | push @args, |
1123 | join("", @{$self->{deferred}}{$first_rec .. $last_rec}), # data |
1124 | $start, # position |
1125 | $end-$start; # length |
1126 | push @adjust, [$first_rec, # starting at this position... |
1127 | $last_rec-$first_rec+1, # this many records... |
1128 | # are replaced with these... |
1129 | @{$self->{deferred}}{$first_rec .. $last_rec}, |
1130 | ]; |
1131 | } |
1132 | |
1133 | $self->_mtwrite(@args); # write multiple record groups |
1134 | $self->_discard; # clear out defered-write-cache |
1135 | $self->_oadjust(@adjust); |
1136 | } |
1137 | |
57c7bc08 |
1138 | # Discard deferred writes and disable future deferred writes |
b3fe5a4c |
1139 | sub discard { |
1140 | my $self = shift; |
57c7bc08 |
1141 | $self->_discard; |
b3fe5a4c |
1142 | $self->{defer} = 0; |
1143 | } |
1144 | |
57c7bc08 |
1145 | # Discard deferred writes, but retain old deferred writing mode |
1146 | sub _discard { |
1147 | my $self = shift; |
6fc0ea7e |
1148 | %{$self->{deferred}} = (); |
1149 | $self->{deferred_s} = 0; |
1150 | $self->{deferred_max} = -1; |
1151 | $self->{cache}->set_limit($self->{memory}); |
1152 | } |
1153 | |
1154 | # Deferred writing is enabled, either explicitly ($self->{defer}) |
1155 | # or automatically ($self->{autodeferring}) |
1156 | sub _is_deferring { |
1157 | my $self = shift; |
1158 | $self->{defer} || $self->{autodeferring}; |
1159 | } |
1160 | |
1161 | # The largest record number of any deferred record |
1162 | sub _defer_max { |
1163 | my $self = shift; |
1164 | return $self->{deferred_max} if defined $self->{deferred_max}; |
1165 | my $max = -1; |
1166 | for my $key (keys %{$self->{deferred}}) { |
1167 | $max = $key if $key > $max; |
1168 | } |
1169 | $self->{deferred_max} = $max; |
1170 | $max; |
57c7bc08 |
1171 | } |
1172 | |
6fc0ea7e |
1173 | ################################################################ |
1174 | # |
1175 | # Matters related to autodeferment |
1176 | # |
1177 | |
1178 | # Get/set autodefer option |
1179 | sub autodefer { |
1180 | my $self = shift; |
1181 | if (@_) { |
1182 | my $old = $self->{autodefer}; |
1183 | $self->{autodefer} = shift; |
1184 | if ($old) { |
1185 | $self->_stop_autodeferring; |
1186 | @{$self->{ad_history}} = (); |
1187 | } |
1188 | $old; |
1189 | } else { |
1190 | $self->{autodefer}; |
1191 | } |
1192 | } |
1193 | |
1194 | # The user is trying to store record #$n Record that in the history, |
1195 | # and then enable (or disable) autodeferment if that seems useful. |
1196 | # Note that it's OK for $n to be a non-number, as long as the function |
1197 | # is prepared to deal with that. Nobody else looks at the ad_history. |
1198 | # |
1199 | # Now, what does the ad_history mean, and what is this function doing? |
1200 | # Essentially, the idea is to enable autodeferring when we see that the |
1201 | # user has made three consecutive STORE calls to three consecutive records. |
1202 | # ("Three" is actually ->{autodefer_threshhold}.) |
1203 | # A STORE call for record #$n inserts $n into the autodefer history, |
1204 | # and if the history contains three consecutive records, we enable |
1205 | # autodeferment. An ad_history of [X, Y] means that the most recent |
1206 | # STOREs were for records X, X+1, ..., Y, in that order. |
1207 | # |
1208 | # Inserting a nonconsecutive number erases the history and starts over. |
1209 | # |
1210 | # Performing a special operation like SPLICE erases the history. |
1211 | # |
1212 | # There's one special case: CLEAR means that CLEAR was just called. |
1213 | # In this case, we prime the history with [-2, -1] so that if the next |
1214 | # write is for record 0, autodeferring goes on immediately. This is for |
1215 | # the common special case of "@a = (...)". |
1216 | # |
1217 | sub _annotate_ad_history { |
1218 | my ($self, $n) = @_; |
1219 | return unless $self->{autodefer}; # feature is disabled |
1220 | return if $self->{defer}; # already in explicit defer mode |
1221 | return unless $self->{offsets}[-1] >= $self->{autodefer_filelen_threshhold}; |
1222 | |
1223 | local *H = $self->{ad_history}; |
1224 | if ($n eq 'CLEAR') { |
1225 | @H = (-2, -1); # prime the history with fake records |
1226 | $self->_stop_autodeferring; |
1227 | } elsif ($n =~ /^\d+$/) { |
1228 | if (@H == 0) { |
1229 | @H = ($n, $n); |
1230 | } else { # @H == 2 |
1231 | if ($H[1] == $n-1) { # another consecutive record |
1232 | $H[1]++; |
1233 | if ($H[1] - $H[0] + 1 >= $self->{autodefer_threshhold}) { |
1234 | $self->{autodeferring} = 1; |
1235 | } |
1236 | } else { # nonconsecutive- erase and start over |
1237 | @H = ($n, $n); |
1238 | $self->_stop_autodeferring; |
1239 | } |
1240 | } |
1241 | } else { # SPLICE or STORESIZE or some such |
1242 | @H = (); |
1243 | $self->_stop_autodeferring; |
1244 | } |
1245 | } |
1246 | |
6ae23f41 |
1247 | # If autodeferring was enabled, cut it out and discard the history |
6fc0ea7e |
1248 | sub _stop_autodeferring { |
1249 | my $self = shift; |
1250 | if ($self->{autodeferring}) { |
1251 | $self->_flush; |
1252 | } |
1253 | $self->{autodeferring} = 0; |
1254 | } |
1255 | |
1256 | ################################################################ |
1257 | |
b3fe5a4c |
1258 | |
57c7bc08 |
1259 | # This is NOT a method. It is here for two reasons: |
1260 | # 1. To factor a fairly complicated block out of the constructor |
1261 | # 2. To provide access for the test suite, which need to be sure |
1262 | # files are being written properly. |
b3fe5a4c |
1263 | sub _default_recsep { |
1264 | my $recsep = $/; |
57c7bc08 |
1265 | if ($^O eq 'MSWin32') { # Dos too? |
b3fe5a4c |
1266 | # Windows users expect files to be terminated with \r\n |
1267 | # But $/ is set to \n instead |
1268 | # Note that this also transforms \n\n into \r\n\r\n. |
1269 | # That is a feature. |
1270 | $recsep =~ s/\n/\r\n/g; |
1271 | } |
1272 | $recsep; |
1273 | } |
1274 | |
57c7bc08 |
1275 | # Utility function for _check_integrity |
1276 | sub _ci_warn { |
1277 | my $msg = shift; |
1278 | $msg =~ s/\n/\\n/g; |
1279 | $msg =~ s/\r/\\r/g; |
1280 | print "# $msg\n"; |
1281 | } |
1282 | |
b5aed31e |
1283 | # Given a file, make sure the cache is consistent with the |
57c7bc08 |
1284 | # file contents and the internal data structures are consistent with |
1285 | # each other. Returns true if everything checks out, false if not |
1286 | # |
1287 | # The $file argument is no longer used. It is retained for compatibility |
1288 | # with the existing test suite. |
b5aed31e |
1289 | sub _check_integrity { |
1290 | my ($self, $file, $warn) = @_; |
6fc0ea7e |
1291 | my $rsl = $self->{recseplen}; |
1292 | my $rs = $self->{recsep}; |
b5aed31e |
1293 | my $good = 1; |
6fc0ea7e |
1294 | local *_; # local $_ does not work here |
1295 | local $DIAGNOSTIC = 1; |
1296 | |
1297 | if (not defined $rs) { |
1298 | _ci_warn("recsep is undef!"); |
1299 | $good = 0; |
1300 | } elsif ($rs eq "") { |
1301 | _ci_warn("recsep is empty!"); |
1302 | $good = 0; |
1303 | } elsif ($rsl != length $rs) { |
1304 | my $ln = length $rs; |
1305 | _ci_warn("recsep <$rs> has length $ln, should be $rsl"); |
1306 | $good = 0; |
1307 | } |
fa408a35 |
1308 | |
836d9961 |
1309 | if (not defined $self->{offsets}[0]) { |
57c7bc08 |
1310 | _ci_warn("offset 0 is missing!"); |
836d9961 |
1311 | $good = 0; |
bf919750 |
1312 | |
836d9961 |
1313 | } elsif ($self->{offsets}[0] != 0) { |
57c7bc08 |
1314 | _ci_warn("rec 0: offset <$self->{offsets}[0]> s/b 0!"); |
b5aed31e |
1315 | $good = 0; |
1316 | } |
fa408a35 |
1317 | |
57c7bc08 |
1318 | my $cached = 0; |
6fc0ea7e |
1319 | { |
1320 | local *F = $self->{fh}; |
1321 | seek F, 0, SEEK_SET; |
1322 | local $. = 0; |
1323 | local $/ = $rs; |
1324 | |
1325 | while (<F>) { |
1326 | my $n = $. - 1; |
1327 | my $cached = $self->{cache}->_produce($n); |
1328 | my $offset = $self->{offsets}[$.]; |
1329 | my $ao = tell F; |
1330 | if (defined $offset && $offset != $ao) { |
1331 | _ci_warn("rec $n: offset <$offset> actual <$ao>"); |
1332 | $good = 0; |
1333 | } |
1334 | if (defined $cached && $_ ne $cached && ! $self->{deferred}{$n}) { |
1335 | $good = 0; |
1336 | _ci_warn("rec $n: cached <$cached> actual <$_>"); |
1337 | } |
1338 | if (defined $cached && substr($cached, -$rsl) ne $rs) { |
27531ffb |
1339 | $good = 0; |
6fc0ea7e |
1340 | _ci_warn("rec $n in the cache is missing the record separator"); |
1341 | } |
27531ffb |
1342 | if (! defined $offset && $self->{eof}) { |
1343 | $good = 0; |
1344 | _ci_warn("The offset table was marked complete, but it is missing element $."); |
1345 | } |
1346 | } |
1347 | if (@{$self->{offsets}} > $.+1) { |
1348 | $good = 0; |
1349 | my $n = @{$self->{offsets}}; |
1350 | _ci_warn("The offset table has $n items, but the file has only $."); |
6fc0ea7e |
1351 | } |
b5aed31e |
1352 | |
6fc0ea7e |
1353 | my $deferring = $self->_is_deferring; |
bf919750 |
1354 | for my $n ($self->{cache}->ckeys) { |
6fc0ea7e |
1355 | my $r = $self->{cache}->_produce($n); |
1356 | $cached += length($r); |
1357 | next if $n+1 <= $.; # checked this already |
1358 | _ci_warn("spurious caching of record $n"); |
b5aed31e |
1359 | $good = 0; |
1360 | } |
6fc0ea7e |
1361 | my $b = $self->{cache}->bytes; |
1362 | if ($cached != $b) { |
1363 | _ci_warn("cache size is $b, should be $cached"); |
b5aed31e |
1364 | $good = 0; |
1365 | } |
1366 | } |
1367 | |
bf919750 |
1368 | # That cache has its own set of tests |
6fc0ea7e |
1369 | $good = 0 unless $self->{cache}->_check_integrity; |
1370 | |
57c7bc08 |
1371 | # Now let's check the deferbuffer |
1372 | # Unless deferred writing is enabled, it should be empty |
6fc0ea7e |
1373 | if (! $self->_is_deferring && %{$self->{deferred}}) { |
57c7bc08 |
1374 | _ci_warn("deferred writing disabled, but deferbuffer nonempty"); |
1375 | $good = 0; |
1376 | } |
1377 | |
1378 | # Any record in the deferbuffer should *not* be present in the readcache |
1379 | my $deferred_s = 0; |
1380 | while (my ($n, $r) = each %{$self->{deferred}}) { |
1381 | $deferred_s += length($r); |
6fc0ea7e |
1382 | if (defined $self->{cache}->_produce($n)) { |
57c7bc08 |
1383 | _ci_warn("record $n is in the deferbuffer *and* the readcache"); |
1384 | $good = 0; |
1385 | } |
6fc0ea7e |
1386 | if (substr($r, -$rsl) ne $rs) { |
57c7bc08 |
1387 | _ci_warn("rec $n in the deferbuffer is missing the record separator"); |
1388 | $good = 0; |
1389 | } |
1390 | } |
1391 | |
1392 | # Total size of deferbuffer should match internal total |
1393 | if ($deferred_s != $self->{deferred_s}) { |
1394 | _ci_warn("buffer size is $self->{deferred_s}, should be $deferred_s"); |
1395 | $good = 0; |
1396 | } |
1397 | |
1398 | # Total size of deferbuffer should not exceed the specified limit |
1399 | if ($deferred_s > $self->{dw_size}) { |
1400 | _ci_warn("buffer size is $self->{deferred_s} which exceeds the limit of $self->{dw_size}"); |
1401 | $good = 0; |
1402 | } |
1403 | |
1404 | # Total size of cached data should not exceed the specified limit |
1405 | if ($deferred_s + $cached > $self->{memory}) { |
1406 | my $total = $deferred_s + $cached; |
1407 | _ci_warn("total stored data size is $total which exceeds the limit of $self->{memory}"); |
1408 | $good = 0; |
1409 | } |
1410 | |
6fc0ea7e |
1411 | # Stuff related to autodeferment |
1412 | if (!$self->{autodefer} && @{$self->{ad_history}}) { |
1413 | _ci_warn("autodefer is disabled, but ad_history is nonempty"); |
1414 | $good = 0; |
1415 | } |
1416 | if ($self->{autodeferring} && $self->{defer}) { |
1417 | _ci_warn("both autodeferring and explicit deferring are active"); |
1418 | $good = 0; |
1419 | } |
1420 | if (@{$self->{ad_history}} == 0) { |
1421 | # That's OK, no additional tests required |
1422 | } elsif (@{$self->{ad_history}} == 2) { |
1423 | my @non_number = grep !/^-?\d+$/, @{$self->{ad_history}}; |
1424 | if (@non_number) { |
1425 | my $msg; |
1426 | { local $" = ')('; |
1427 | $msg = "ad_history contains non-numbers (@{$self->{ad_history}})"; |
1428 | } |
1429 | _ci_warn($msg); |
1430 | $good = 0; |
1431 | } elsif ($self->{ad_history}[1] < $self->{ad_history}[0]) { |
1432 | _ci_warn("ad_history has nonsensical values @{$self->{ad_history}}"); |
1433 | $good = 0; |
1434 | } |
1435 | } else { |
1436 | _ci_warn("ad_history has bad length <@{$self->{ad_history}}>"); |
1437 | $good = 0; |
1438 | } |
1439 | |
b5aed31e |
1440 | $good; |
1441 | } |
1442 | |
6fc0ea7e |
1443 | ################################################################ |
1444 | # |
1445 | # Tie::File::Cache |
1446 | # |
1447 | # Read cache |
1448 | |
1449 | package Tie::File::Cache; |
1450 | $Tie::File::Cache::VERSION = $Tie::File::VERSION; |
1451 | use Carp ':DEFAULT', 'confess'; |
1452 | |
1453 | sub HEAP () { 0 } |
1454 | sub HASH () { 1 } |
1455 | sub MAX () { 2 } |
1456 | sub BYTES() { 3 } |
6ae23f41 |
1457 | #sub STAT () { 4 } # Array with request statistics for each record |
1458 | #sub MISS () { 5 } # Total number of cache misses |
1459 | #sub REQ () { 6 } # Total number of cache requests |
6fc0ea7e |
1460 | use strict 'vars'; |
1461 | |
1462 | sub new { |
1463 | my ($pack, $max) = @_; |
1464 | local *_; |
1465 | croak "missing argument to ->new" unless defined $max; |
1466 | my $self = []; |
1467 | bless $self => $pack; |
1468 | @$self = (Tie::File::Heap->new($self), {}, $max, 0); |
1469 | $self; |
1470 | } |
1471 | |
1472 | sub adj_limit { |
1473 | my ($self, $n) = @_; |
1474 | $self->[MAX] += $n; |
1475 | } |
1476 | |
1477 | sub set_limit { |
1478 | my ($self, $n) = @_; |
1479 | $self->[MAX] = $n; |
1480 | } |
1481 | |
1482 | # For internal use only |
1483 | # Will be called by the heap structure to notify us that a certain |
1484 | # piece of data has moved from one heap element to another. |
1485 | # $k is the hash key of the item |
1486 | # $n is the new index into the heap at which it is stored |
1487 | # If $n is undefined, the item has been removed from the heap. |
1488 | sub _heap_move { |
1489 | my ($self, $k, $n) = @_; |
1490 | if (defined $n) { |
1491 | $self->[HASH]{$k} = $n; |
1492 | } else { |
6ae23f41 |
1493 | delete $self->[HASH]{$k}; |
6fc0ea7e |
1494 | } |
1495 | } |
1496 | |
1497 | sub insert { |
1498 | my ($self, $key, $val) = @_; |
1499 | local *_; |
1500 | croak "missing argument to ->insert" unless defined $key; |
1501 | unless (defined $self->[MAX]) { |
1502 | confess "undefined max" ; |
1503 | } |
1504 | confess "undefined val" unless defined $val; |
1505 | return if length($val) > $self->[MAX]; |
6ae23f41 |
1506 | |
1507 | # if ($self->[STAT]) { |
1508 | # $self->[STAT][$key] = 1; |
1509 | # return; |
1510 | # } |
1511 | |
6fc0ea7e |
1512 | my $oldnode = $self->[HASH]{$key}; |
1513 | if (defined $oldnode) { |
1514 | my $oldval = $self->[HEAP]->set_val($oldnode, $val); |
1515 | $self->[BYTES] -= length($oldval); |
1516 | } else { |
1517 | $self->[HEAP]->insert($key, $val); |
1518 | } |
1519 | $self->[BYTES] += length($val); |
6ae23f41 |
1520 | $self->flush if $self->[BYTES] > $self->[MAX]; |
6fc0ea7e |
1521 | } |
1522 | |
1523 | sub expire { |
1524 | my $self = shift; |
1525 | my $old_data = $self->[HEAP]->popheap; |
1526 | return unless defined $old_data; |
1527 | $self->[BYTES] -= length $old_data; |
1528 | $old_data; |
1529 | } |
1530 | |
1531 | sub remove { |
1532 | my ($self, @keys) = @_; |
1533 | my @result; |
6ae23f41 |
1534 | |
1535 | # if ($self->[STAT]) { |
1536 | # for my $key (@keys) { |
1537 | # $self->[STAT][$key] = 0; |
1538 | # } |
1539 | # return; |
1540 | # } |
1541 | |
6fc0ea7e |
1542 | for my $key (@keys) { |
1543 | next unless exists $self->[HASH]{$key}; |
1544 | my $old_data = $self->[HEAP]->remove($self->[HASH]{$key}); |
1545 | $self->[BYTES] -= length $old_data; |
1546 | push @result, $old_data; |
1547 | } |
1548 | @result; |
1549 | } |
1550 | |
1551 | sub lookup { |
1552 | my ($self, $key) = @_; |
1553 | local *_; |
1554 | croak "missing argument to ->lookup" unless defined $key; |
6ae23f41 |
1555 | |
1556 | # if ($self->[STAT]) { |
1557 | # $self->[MISS]++ if $self->[STAT][$key]++ == 0; |
1558 | # $self->[REQ]++; |
1559 | # my $hit_rate = 1 - $self->[MISS] / $self->[REQ]; |
1560 | # # Do some testing to determine this threshhold |
1561 | # $#$self = STAT - 1 if $hit_rate > 0.20; |
1562 | # } |
1563 | |
6fc0ea7e |
1564 | if (exists $self->[HASH]{$key}) { |
1565 | $self->[HEAP]->lookup($self->[HASH]{$key}); |
1566 | } else { |
1567 | return; |
1568 | } |
1569 | } |
1570 | |
1571 | # For internal use only |
1572 | sub _produce { |
1573 | my ($self, $key) = @_; |
1574 | my $loc = $self->[HASH]{$key}; |
1575 | return unless defined $loc; |
1576 | $self->[HEAP][$loc][2]; |
1577 | } |
1578 | |
1579 | # For internal use only |
1580 | sub _promote { |
1581 | my ($self, $key) = @_; |
1582 | $self->[HEAP]->promote($self->[HASH]{$key}); |
1583 | } |
1584 | |
1585 | sub empty { |
1586 | my ($self) = @_; |
1587 | %{$self->[HASH]} = (); |
1588 | $self->[BYTES] = 0; |
1589 | $self->[HEAP]->empty; |
6ae23f41 |
1590 | # @{$self->[STAT]} = (); |
1591 | # $self->[MISS] = 0; |
1592 | # $self->[REQ] = 0; |
6fc0ea7e |
1593 | } |
1594 | |
1595 | sub is_empty { |
1596 | my ($self) = @_; |
1597 | keys %{$self->[HASH]} == 0; |
1598 | } |
1599 | |
1600 | sub update { |
1601 | my ($self, $key, $val) = @_; |
1602 | local *_; |
1603 | croak "missing argument to ->update" unless defined $key; |
1604 | if (length($val) > $self->[MAX]) { |
6ae23f41 |
1605 | my ($oldval) = $self->remove($key); |
6fc0ea7e |
1606 | $self->[BYTES] -= length($oldval) if defined $oldval; |
1607 | } elsif (exists $self->[HASH]{$key}) { |
1608 | my $oldval = $self->[HEAP]->set_val($self->[HASH]{$key}, $val); |
1609 | $self->[BYTES] += length($val); |
1610 | $self->[BYTES] -= length($oldval) if defined $oldval; |
1611 | } else { |
1612 | $self->[HEAP]->insert($key, $val); |
1613 | $self->[BYTES] += length($val); |
1614 | } |
1615 | $self->flush; |
1616 | } |
1617 | |
1618 | sub rekey { |
1619 | my ($self, $okeys, $nkeys) = @_; |
1620 | local *_; |
1621 | my %map; |
1622 | @map{@$okeys} = @$nkeys; |
1623 | croak "missing argument to ->rekey" unless defined $nkeys; |
1624 | croak "length mismatch in ->rekey arguments" unless @$nkeys == @$okeys; |
1625 | my %adjusted; # map new keys to heap indices |
1626 | # You should be able to cut this to one loop TODO XXX |
1627 | for (0 .. $#$okeys) { |
1628 | $adjusted{$nkeys->[$_]} = delete $self->[HASH]{$okeys->[$_]}; |
1629 | } |
1630 | while (my ($nk, $ix) = each %adjusted) { |
1631 | # @{$self->[HASH]}{keys %adjusted} = values %adjusted; |
1632 | $self->[HEAP]->rekey($ix, $nk); |
1633 | $self->[HASH]{$nk} = $ix; |
1634 | } |
1635 | } |
1636 | |
bf919750 |
1637 | sub ckeys { |
6fc0ea7e |
1638 | my $self = shift; |
1639 | my @a = keys %{$self->[HASH]}; |
1640 | @a; |
1641 | } |
1642 | |
6ae23f41 |
1643 | # Return total amount of cached data |
6fc0ea7e |
1644 | sub bytes { |
1645 | my $self = shift; |
1646 | $self->[BYTES]; |
1647 | } |
1648 | |
6ae23f41 |
1649 | # Expire oldest item from cache until cache size is smaller than $max |
6fc0ea7e |
1650 | sub reduce_size_to { |
1651 | my ($self, $max) = @_; |
6ae23f41 |
1652 | until ($self->[BYTES] <= $max) { |
1653 | # Note that Tie::File::Cache::expire has been inlined here |
1654 | my $old_data = $self->[HEAP]->popheap; |
1655 | return unless defined $old_data; |
1656 | $self->[BYTES] -= length $old_data; |
6fc0ea7e |
1657 | } |
1658 | } |
1659 | |
6ae23f41 |
1660 | # Why not just $self->reduce_size_to($self->[MAX])? |
1661 | # Try this when things stabilize TODO XXX |
1662 | # If the cache is too full, expire the oldest records |
6fc0ea7e |
1663 | sub flush { |
1664 | my $self = shift; |
6ae23f41 |
1665 | $self->reduce_size_to($self->[MAX]) if $self->[BYTES] > $self->[MAX]; |
6fc0ea7e |
1666 | } |
1667 | |
1668 | # For internal use only |
1669 | sub _produce_lru { |
1670 | my $self = shift; |
1671 | $self->[HEAP]->expire_order; |
1672 | } |
1673 | |
bf919750 |
1674 | BEGIN { *_ci_warn = \&Tie::File::_ci_warn } |
1675 | |
1676 | sub _check_integrity { # For CACHE |
6fc0ea7e |
1677 | my $self = shift; |
bf919750 |
1678 | my $good = 1; |
1679 | |
1680 | # Test HEAP |
1681 | $self->[HEAP]->_check_integrity or $good = 0; |
1682 | |
1683 | # Test HASH |
1684 | my $bytes = 0; |
1685 | for my $k (keys %{$self->[HASH]}) { |
1686 | if ($k ne '0' && $k !~ /^[1-9][0-9]*$/) { |
1687 | $good = 0; |
1688 | _ci_warn "Cache hash key <$k> is non-numeric"; |
1689 | } |
1690 | |
1691 | my $h = $self->[HASH]{$k}; |
1692 | if (! defined $h) { |
1693 | $good = 0; |
1694 | _ci_warn "Heap index number for key $k is undefined"; |
1695 | } elsif ($h == 0) { |
1696 | $good = 0; |
1697 | _ci_warn "Heap index number for key $k is zero"; |
1698 | } else { |
1699 | my $j = $self->[HEAP][$h]; |
1700 | if (! defined $j) { |
1701 | $good = 0; |
1702 | _ci_warn "Heap contents key $k (=> $h) are undefined"; |
1703 | } else { |
1704 | $bytes += length($j->[2]); |
1705 | if ($k ne $j->[1]) { |
1706 | $good = 0; |
1707 | _ci_warn "Heap contents key $k (=> $h) is $j->[1], should be $k"; |
1708 | } |
1709 | } |
1710 | } |
1711 | } |
1712 | |
1713 | # Test BYTES |
1714 | if ($bytes != $self->[BYTES]) { |
1715 | $good = 0; |
1716 | _ci_warn "Total data in cache is $bytes, expected $self->[BYTES]"; |
1717 | } |
1718 | |
1719 | # Test MAX |
1720 | if ($bytes > $self->[MAX]) { |
1721 | $good = 0; |
1722 | _ci_warn "Total data in cache is $bytes, exceeds maximum $self->[MAX]"; |
1723 | } |
1724 | |
1725 | return $good; |
6fc0ea7e |
1726 | } |
1727 | |
1728 | sub delink { |
1729 | my $self = shift; |
1730 | $self->[HEAP] = undef; # Bye bye heap |
1731 | } |
1732 | |
1733 | ################################################################ |
1734 | # |
1735 | # Tie::File::Heap |
1736 | # |
1737 | # Heap data structure for use by cache LRU routines |
1738 | |
1739 | package Tie::File::Heap; |
1740 | use Carp ':DEFAULT', 'confess'; |
1741 | $Tie::File::Heap::VERSION = $Tie::File::Cache::VERSION; |
1742 | sub SEQ () { 0 }; |
1743 | sub KEY () { 1 }; |
1744 | sub DAT () { 2 }; |
1745 | |
1746 | sub new { |
1747 | my ($pack, $cache) = @_; |
1748 | die "$pack: Parent cache object $cache does not support _heap_move method" |
1749 | unless eval { $cache->can('_heap_move') }; |
1750 | my $self = [[0,$cache,0]]; |
1751 | bless $self => $pack; |
1752 | } |
1753 | |
1754 | # Allocate a new sequence number, larger than all previously allocated numbers |
1755 | sub _nseq { |
1756 | my $self = shift; |
1757 | $self->[0][0]++; |
1758 | } |
1759 | |
1760 | sub _cache { |
1761 | my $self = shift; |
1762 | $self->[0][1]; |
1763 | } |
1764 | |
1765 | sub _nelts { |
1766 | my $self = shift; |
1767 | $self->[0][2]; |
1768 | } |
1769 | |
1770 | sub _nelts_inc { |
1771 | my $self = shift; |
1772 | ++$self->[0][2]; |
1773 | } |
1774 | |
1775 | sub _nelts_dec { |
1776 | my $self = shift; |
1777 | --$self->[0][2]; |
1778 | } |
1779 | |
1780 | sub is_empty { |
1781 | my $self = shift; |
1782 | $self->_nelts == 0; |
1783 | } |
1784 | |
1785 | sub empty { |
1786 | my $self = shift; |
1787 | $#$self = 0; |
1788 | $self->[0][2] = 0; |
1789 | $self->[0][0] = 0; # might as well reset the sequence numbers |
1790 | } |
1791 | |
27531ffb |
1792 | # notify the parent cache object that we moved something |
6fc0ea7e |
1793 | sub _heap_move { |
1794 | my $self = shift; |
1795 | $self->_cache->_heap_move(@_); |
1796 | } |
1797 | |
1798 | # Insert a piece of data into the heap with the indicated sequence number. |
1799 | # The item with the smallest sequence number is always at the top. |
1800 | # If no sequence number is specified, allocate a new one and insert the |
1801 | # item at the bottom. |
1802 | sub insert { |
1803 | my ($self, $key, $data, $seq) = @_; |
1804 | $seq = $self->_nseq unless defined $seq; |
1805 | $self->_insert_new([$seq, $key, $data]); |
1806 | } |
1807 | |
1808 | # Insert a new, fresh item at the bottom of the heap |
1809 | sub _insert_new { |
1810 | my ($self, $item) = @_; |
1811 | my $i = @$self; |
1812 | $i = int($i/2) until defined $self->[$i/2]; |
1813 | $self->[$i] = $item; |
27531ffb |
1814 | $self->[0][1]->_heap_move($self->[$i][KEY], $i); |
6fc0ea7e |
1815 | $self->_nelts_inc; |
1816 | } |
1817 | |
1818 | # Insert [$data, $seq] pair at or below item $i in the heap. |
1819 | # If $i is omitted, default to 1 (the top element.) |
1820 | sub _insert { |
1821 | my ($self, $item, $i) = @_; |
bf919750 |
1822 | # $self->_check_loc($i) if defined $i; |
6fc0ea7e |
1823 | $i = 1 unless defined $i; |
1824 | until (! defined $self->[$i]) { |
1825 | if ($self->[$i][SEQ] > $item->[SEQ]) { # inserted item is older |
1826 | ($self->[$i], $item) = ($item, $self->[$i]); |
27531ffb |
1827 | $self->[0][1]->_heap_move($self->[$i][KEY], $i); |
6fc0ea7e |
1828 | } |
1829 | # If either is undefined, go that way. Otherwise, choose at random |
1830 | my $dir; |
1831 | $dir = 0 if !defined $self->[2*$i]; |
1832 | $dir = 1 if !defined $self->[2*$i+1]; |
1833 | $dir = int(rand(2)) unless defined $dir; |
1834 | $i = 2*$i + $dir; |
1835 | } |
1836 | $self->[$i] = $item; |
27531ffb |
1837 | $self->[0][1]->_heap_move($self->[$i][KEY], $i); |
6fc0ea7e |
1838 | $self->_nelts_inc; |
1839 | } |
1840 | |
1841 | # Remove the item at node $i from the heap, moving child items upwards. |
1842 | # The item with the smallest sequence number is always at the top. |
1843 | # Moving items upwards maintains this condition. |
6ae23f41 |
1844 | # Return the removed item. Return undef if there was no item at node $i. |
6fc0ea7e |
1845 | sub remove { |
1846 | my ($self, $i) = @_; |
1847 | $i = 1 unless defined $i; |
1848 | my $top = $self->[$i]; |
1849 | return unless defined $top; |
1850 | while (1) { |
1851 | my $ii; |
1852 | my ($L, $R) = (2*$i, 2*$i+1); |
1853 | |
1854 | # If either is undefined, go the other way. |
1855 | # Otherwise, go towards the smallest. |
1856 | last unless defined $self->[$L] || defined $self->[$R]; |
1857 | $ii = $R if not defined $self->[$L]; |
1858 | $ii = $L if not defined $self->[$R]; |
1859 | unless (defined $ii) { |
1860 | $ii = $self->[$L][SEQ] < $self->[$R][SEQ] ? $L : $R; |
1861 | } |
1862 | |
1863 | $self->[$i] = $self->[$ii]; # Promote child to fill vacated spot |
27531ffb |
1864 | $self->[0][1]->_heap_move($self->[$i][KEY], $i); |
6fc0ea7e |
1865 | $i = $ii; # Fill new vacated spot |
1866 | } |
27531ffb |
1867 | $self->[0][1]->_heap_move($top->[KEY], undef); |
6fc0ea7e |
1868 | undef $self->[$i]; |
1869 | $self->_nelts_dec; |
1870 | return $top->[DAT]; |
1871 | } |
1872 | |
1873 | sub popheap { |
1874 | my $self = shift; |
1875 | $self->remove(1); |
1876 | } |
1877 | |
1878 | # set the sequence number of the indicated item to a higher number |
1879 | # than any other item in the heap, and bubble the item down to the |
1880 | # bottom. |
1881 | sub promote { |
1882 | my ($self, $n) = @_; |
bf919750 |
1883 | # $self->_check_loc($n); |
6fc0ea7e |
1884 | $self->[$n][SEQ] = $self->_nseq; |
1885 | my $i = $n; |
1886 | while (1) { |
1887 | my ($L, $R) = (2*$i, 2*$i+1); |
1888 | my $dir; |
1889 | last unless defined $self->[$L] || defined $self->[$R]; |
1890 | $dir = $R unless defined $self->[$L]; |
1891 | $dir = $L unless defined $self->[$R]; |
1892 | unless (defined $dir) { |
1893 | $dir = $self->[$L][SEQ] < $self->[$R][SEQ] ? $L : $R; |
1894 | } |
1895 | @{$self}[$i, $dir] = @{$self}[$dir, $i]; |
1896 | for ($i, $dir) { |
27531ffb |
1897 | $self->[0][1]->_heap_move($self->[$_][KEY], $_) if defined $self->[$_]; |
6fc0ea7e |
1898 | } |
1899 | $i = $dir; |
1900 | } |
1901 | } |
1902 | |
1903 | # Return item $n from the heap, promoting its LRU status |
1904 | sub lookup { |
1905 | my ($self, $n) = @_; |
bf919750 |
1906 | # $self->_check_loc($n); |
6fc0ea7e |
1907 | my $val = $self->[$n]; |
1908 | $self->promote($n); |
1909 | $val->[DAT]; |
1910 | } |
1911 | |
1912 | |
1913 | # Assign a new value for node $n, promoting it to the bottom of the heap |
1914 | sub set_val { |
1915 | my ($self, $n, $val) = @_; |
bf919750 |
1916 | # $self->_check_loc($n); |
6fc0ea7e |
1917 | my $oval = $self->[$n][DAT]; |
1918 | $self->[$n][DAT] = $val; |
1919 | $self->promote($n); |
1920 | return $oval; |
1921 | } |
1922 | |
1923 | # The hask key has changed for an item; |
1924 | # alter the heap's record of the hash key |
1925 | sub rekey { |
1926 | my ($self, $n, $new_key) = @_; |
bf919750 |
1927 | # $self->_check_loc($n); |
6fc0ea7e |
1928 | $self->[$n][KEY] = $new_key; |
1929 | } |
1930 | |
1931 | sub _check_loc { |
1932 | my ($self, $n) = @_; |
bf919750 |
1933 | unless (1 || defined $self->[$n]) { |
6fc0ea7e |
1934 | confess "_check_loc($n) failed"; |
1935 | } |
1936 | } |
1937 | |
bf919750 |
1938 | BEGIN { *_ci_warn = \&Tie::File::_ci_warn } |
1939 | |
6fc0ea7e |
1940 | sub _check_integrity { |
1941 | my $self = shift; |
1942 | my $good = 1; |
bf919750 |
1943 | my %seq; |
1944 | |
6fc0ea7e |
1945 | unless (eval {$self->[0][1]->isa("Tie::File::Cache")}) { |
bf919750 |
1946 | _ci_warn "Element 0 of heap corrupt"; |
6fc0ea7e |
1947 | $good = 0; |
1948 | } |
1949 | $good = 0 unless $self->_satisfies_heap_condition(1); |
1950 | for my $i (2 .. $#{$self}) { |
1951 | my $p = int($i/2); # index of parent node |
1952 | if (defined $self->[$i] && ! defined $self->[$p]) { |
bf919750 |
1953 | _ci_warn "Element $i of heap defined, but parent $p isn't"; |
6fc0ea7e |
1954 | $good = 0; |
1955 | } |
bf919750 |
1956 | |
1957 | if (defined $self->[$i]) { |
1958 | if ($seq{$self->[$i][SEQ]}) { |
1959 | my $seq = $self->[$i][SEQ]; |
1960 | _ci_warn "Nodes $i and $seq{$seq} both have SEQ=$seq"; |
1961 | $good = 0; |
1962 | } else { |
1963 | $seq{$self->[$i][SEQ]} = $i; |
1964 | } |
1965 | } |
6fc0ea7e |
1966 | } |
bf919750 |
1967 | |
6fc0ea7e |
1968 | return $good; |
1969 | } |
1970 | |
1971 | sub _satisfies_heap_condition { |
1972 | my $self = shift; |
1973 | my $n = shift || 1; |
1974 | my $good = 1; |
1975 | for (0, 1) { |
1976 | my $c = $n*2 + $_; |
1977 | next unless defined $self->[$c]; |
1978 | if ($self->[$n][SEQ] >= $self->[$c]) { |
bf919750 |
1979 | _ci_warn "Node $n of heap does not predate node $c"; |
6fc0ea7e |
1980 | $good = 0 ; |
1981 | } |
1982 | $good = 0 unless $self->_satisfies_heap_condition($c); |
1983 | } |
1984 | return $good; |
1985 | } |
1986 | |
1987 | # Return a list of all the values, sorted by expiration order |
1988 | sub expire_order { |
1989 | my $self = shift; |
1990 | my @nodes = sort {$a->[SEQ] <=> $b->[SEQ]} $self->_nodes; |
1991 | map { $_->[KEY] } @nodes; |
1992 | } |
1993 | |
1994 | sub _nodes { |
1995 | my $self = shift; |
1996 | my $i = shift || 1; |
1997 | return unless defined $self->[$i]; |
1998 | ($self->[$i], $self->_nodes($i*2), $self->_nodes($i*2+1)); |
1999 | } |
2000 | |
fa408a35 |
2001 | "Cogito, ergo sum."; # don't forget to return a true value from the file |
2002 | |
6ae23f41 |
2003 | __END__ |
2004 | |
b5aed31e |
2005 | =head1 NAME |
2006 | |
2007 | Tie::File - Access the lines of a disk file via a Perl array |
2008 | |
2009 | =head1 SYNOPSIS |
2010 | |
bc2f65e0 |
2011 | # This file documents Tie::File version 0.97 |
6ae23f41 |
2012 | use Tie::File; |
b5aed31e |
2013 | |
2014 | tie @array, 'Tie::File', filename or die ...; |
2015 | |
2016 | $array[13] = 'blah'; # line 13 of the file is now 'blah' |
2017 | print $array[42]; # display line 42 of the file |
2018 | |
2019 | $n_recs = @array; # how many records are in the file? |
57c7bc08 |
2020 | $#array -= 2; # chop two records off the end |
2021 | |
b5aed31e |
2022 | |
57c7bc08 |
2023 | for (@array) { |
2024 | s/PERL/Perl/g; # Replace PERL with Perl everywhere in the file |
2025 | } |
2026 | |
2027 | # These are just like regular push, pop, unshift, shift, and splice |
2028 | # Except that they modify the file in the way you would expect |
51efdd02 |
2029 | |
2030 | push @array, new recs...; |
2031 | my $r1 = pop @array; |
2032 | unshift @array, new recs...; |
6ae23f41 |
2033 | my $r2 = shift @array; |
b5aed31e |
2034 | @old_recs = splice @array, 3, 7, new recs...; |
2035 | |
2036 | untie @array; # all finished |
2037 | |
57c7bc08 |
2038 | |
b5aed31e |
2039 | =head1 DESCRIPTION |
2040 | |
2041 | C<Tie::File> represents a regular text file as a Perl array. Each |
2042 | element in the array corresponds to a record in the file. The first |
2043 | line of the file is element 0 of the array; the second line is element |
2044 | 1, and so on. |
2045 | |
2046 | The file is I<not> loaded into memory, so this will work even for |
2047 | gigantic files. |
2048 | |
2049 | Changes to the array are reflected in the file immediately. |
2050 | |
57c7bc08 |
2051 | Lazy people and beginners may now stop reading the manual. |
b3fe5a4c |
2052 | |
b5aed31e |
2053 | =head2 C<recsep> |
2054 | |
2055 | What is a 'record'? By default, the meaning is the same as for the |
2056 | C<E<lt>...E<gt>> operator: It's a string terminated by C<$/>, which is |
6ae23f41 |
2057 | probably C<"\n">. (Minor exception: on DOS and Win32 systems, a |
b3fe5a4c |
2058 | 'record' is a string terminated by C<"\r\n">.) You may change the |
2059 | definition of "record" by supplying the C<recsep> option in the C<tie> |
2060 | call: |
b5aed31e |
2061 | |
2062 | tie @array, 'Tie::File', $file, recsep => 'es'; |
2063 | |
b3fe5a4c |
2064 | This says that records are delimited by the string C<es>. If the file |
2065 | contained the following data: |
b5aed31e |
2066 | |
2067 | Curse these pesky flies!\n |
2068 | |
27531ffb |
2069 | then the C<@array> would appear to have four elements: |
b5aed31e |
2070 | |
0b28bc9a |
2071 | "Curse th" |
2072 | "e p" |
2073 | "ky fli" |
b5aed31e |
2074 | "!\n" |
2075 | |
2076 | An undefined value is not permitted as a record separator. Perl's |
2077 | special "paragraph mode" semantics (E<agrave> la C<$/ = "">) are not |
2078 | emulated. |
2079 | |
0b28bc9a |
2080 | Records read from the tied array do not have the record separator |
27531ffb |
2081 | string on the end; this is to allow |
0b28bc9a |
2082 | |
2083 | $array[17] .= "extra"; |
2084 | |
2085 | to work as expected. |
2086 | |
2087 | (See L<"autochomp">, below.) Records stored into the array will have |
2088 | the record separator string appended before they are written to the |
2089 | file, if they don't have one already. For example, if the record |
2090 | separator string is C<"\n">, then the following two lines do exactly |
2091 | the same thing: |
b5aed31e |
2092 | |
2093 | $array[17] = "Cherry pie"; |
2094 | $array[17] = "Cherry pie\n"; |
2095 | |
2096 | The result is that the contents of line 17 of the file will be |
2097 | replaced with "Cherry pie"; a newline character will separate line 17 |
27531ffb |
2098 | from line 18. This means that this code will do nothing: |
b5aed31e |
2099 | |
2100 | chomp $array[17]; |
2101 | |
2102 | Because the C<chomp>ed value will have the separator reattached when |
2103 | it is written back to the file. There is no way to create a file |
2104 | whose trailing record separator string is missing. |
2105 | |
27531ffb |
2106 | Inserting records that I<contain> the record separator string is not |
2107 | supported by this module. It will probably produce a reasonable |
2108 | result, but what this result will be may change in a future version. |
2109 | Use 'splice' to insert records or to replace one record with several. |
b5aed31e |
2110 | |
0b28bc9a |
2111 | =head2 C<autochomp> |
2112 | |
2113 | Normally, array elements have the record separator removed, so that if |
2114 | the file contains the text |
2115 | |
2116 | Gold |
2117 | Frankincense |
2118 | Myrrh |
2119 | |
57c7bc08 |
2120 | the tied array will appear to contain C<("Gold", "Frankincense", |
2121 | "Myrrh")>. If you set C<autochomp> to a false value, the record |
2122 | separator will not be removed. If the file above was tied with |
0b28bc9a |
2123 | |
2124 | tie @gifts, "Tie::File", $gifts, autochomp => 0; |
2125 | |
2126 | then the array C<@gifts> would appear to contain C<("Gold\n", |
2127 | "Frankincense\n", "Myrrh\n")>, or (on Win32 systems) C<("Gold\r\n", |
2128 | "Frankincense\r\n", "Myrrh\r\n")>. |
2129 | |
b5aed31e |
2130 | =head2 C<mode> |
2131 | |
2132 | Normally, the specified file will be opened for read and write access, |
2133 | and will be created if it does not exist. (That is, the flags |
2134 | C<O_RDWR | O_CREAT> are supplied in the C<open> call.) If you want to |
2135 | change this, you may supply alternative flags in the C<mode> option. |
2136 | See L<Fcntl> for a listing of available flags. |
2137 | For example: |
2138 | |
2139 | # open the file if it exists, but fail if it does not exist |
2140 | use Fcntl 'O_RDWR'; |
2141 | tie @array, 'Tie::File', $file, mode => O_RDWR; |
2142 | |
2143 | # create the file if it does not exist |
2144 | use Fcntl 'O_RDWR', 'O_CREAT'; |
2145 | tie @array, 'Tie::File', $file, mode => O_RDWR | O_CREAT; |
2146 | |
2147 | # open an existing file in read-only mode |
2148 | use Fcntl 'O_RDONLY'; |
2149 | tie @array, 'Tie::File', $file, mode => O_RDONLY; |
2150 | |
2151 | Opening the data file in write-only or append mode is not supported. |
2152 | |
b3fe5a4c |
2153 | =head2 C<memory> |
2154 | |
57c7bc08 |
2155 | This is an upper limit on the amount of memory that C<Tie::File> will |
2156 | consume at any time while managing the file. This is used for two |
2157 | things: managing the I<read cache> and managing the I<deferred write |
2158 | buffer>. |
b5aed31e |
2159 | |
2160 | Records read in from the file are cached, to avoid having to re-read |
2161 | them repeatedly. If you read the same record twice, the first time it |
2162 | will be stored in memory, and the second time it will be fetched from |
b3fe5a4c |
2163 | the I<read cache>. The amount of data in the read cache will not |
2164 | exceed the value you specified for C<memory>. If C<Tie::File> wants |
2165 | to cache a new record, but the read cache is full, it will make room |
2166 | by expiring the least-recently visited records from the read cache. |
b5aed31e |
2167 | |
b3fe5a4c |
2168 | The default memory limit is 2Mib. You can adjust the maximum read |
2169 | cache size by supplying the C<memory> option. The argument is the |
2170 | desired cache size, in bytes. |
b5aed31e |
2171 | |
2172 | # I have a lot of memory, so use a large cache to speed up access |
b3fe5a4c |
2173 | tie @array, 'Tie::File', $file, memory => 20_000_000; |
b5aed31e |
2174 | |
b3fe5a4c |
2175 | Setting the memory limit to 0 will inhibit caching; records will be |
b5aed31e |
2176 | fetched from disk every time you examine them. |
2177 | |
27531ffb |
2178 | The C<memory> value is not an absolute or exact limit on the memory |
2179 | used. C<Tie::File> objects contains some structures besides the read |
2180 | cache and the deferred write buffer, whose sizes are not charged |
6ae23f41 |
2181 | against C<memory>. |
2182 | |
2183 | The cache itself consumes about 310 bytes per cached record, so if |
2184 | your file has many short records, you may want to decrease the cache |
2185 | memory limit, or else the cache overhead may exceed the size of the |
2186 | cached data. |
2187 | |
27531ffb |
2188 | |
57c7bc08 |
2189 | =head2 C<dw_size> |
2190 | |
2191 | (This is an advanced feature. Skip this section on first reading.) |
27531ffb |
2192 | |
57c7bc08 |
2193 | If you use deferred writing (See L<"Deferred Writing">, below) then |
2194 | data you write into the array will not be written directly to the |
2195 | file; instead, it will be saved in the I<deferred write buffer> to be |
2196 | written out later. Data in the deferred write buffer is also charged |
2197 | against the memory limit you set with the C<memory> option. |
2198 | |
2199 | You may set the C<dw_size> option to limit the amount of data that can |
2200 | be saved in the deferred write buffer. This limit may not exceed the |
2201 | total memory limit. For example, if you set C<dw_size> to 1000 and |
2202 | C<memory> to 2500, that means that no more than 1000 bytes of deferred |
2203 | writes will be saved up. The space available for the read cache will |
2204 | vary, but it will always be at least 1500 bytes (if the deferred write |
2205 | buffer is full) and it could grow as large as 2500 bytes (if the |
2206 | deferred write buffer is empty.) |
2207 | |
2208 | If you don't specify a C<dw_size>, it defaults to the entire memory |
2209 | limit. |
2210 | |
b5aed31e |
2211 | =head2 Option Format |
2212 | |
2213 | C<-mode> is a synonym for C<mode>. C<-recsep> is a synonym for |
b3fe5a4c |
2214 | C<recsep>. C<-memory> is a synonym for C<memory>. You get the |
b5aed31e |
2215 | idea. |
2216 | |
2217 | =head1 Public Methods |
2218 | |
27531ffb |
2219 | The C<tie> call returns an object, say C<$o>. You may call |
b5aed31e |
2220 | |
2221 | $rec = $o->FETCH($n); |
2222 | $o->STORE($n, $rec); |
2223 | |
b3fe5a4c |
2224 | to fetch or store the record at line C<$n>, respectively; similarly |
2225 | the other tied array methods. (See L<perltie> for details.) You may |
2226 | also call the following methods on this object: |
51efdd02 |
2227 | |
2228 | =head2 C<flock> |
2229 | |
2230 | $o->flock(MODE) |
2231 | |
2232 | will lock the tied file. C<MODE> has the same meaning as the second |
2233 | argument to the Perl built-in C<flock> function; for example |
2234 | C<LOCK_SH> or C<LOCK_EX | LOCK_NB>. (These constants are provided by |
2235 | the C<use Fcntl ':flock'> declaration.) |
2236 | |
57c7bc08 |
2237 | C<MODE> is optional; the default is C<LOCK_EX>. |
2238 | |
6ae23f41 |
2239 | C<Tie::File> maintains an internal table of the byte offset of each |
2240 | record it has seen in the file. |
2241 | |
2242 | When you use C<flock> to lock the file, C<Tie::File> assumes that the |
2243 | read cache is no longer trustworthy, because another process might |
2244 | have modified the file since the last time it was read. Therefore, a |
2245 | successful call to C<flock> discards the contents of the read cache |
2246 | and the internal record offset table. |
2247 | |
57c7bc08 |
2248 | C<Tie::File> promises that the following sequence of operations will |
2249 | be safe: |
2250 | |
2251 | my $o = tie @array, "Tie::File", $filename; |
2252 | $o->flock; |
2253 | |
2254 | In particular, C<Tie::File> will I<not> read or write the file during |
2255 | the C<tie> call. (Exception: Using C<mode =E<gt> O_TRUNC> will, of |
2256 | course, erase the file during the C<tie> call. If you want to do this |
2257 | safely, then open the file without C<O_TRUNC>, lock the file, and use |
2258 | C<@array = ()>.) |
51efdd02 |
2259 | |
2260 | The best way to unlock a file is to discard the object and untie the |
2261 | array. It is probably unsafe to unlock the file without also untying |
2262 | it, because if you do, changes may remain unwritten inside the object. |
2263 | That is why there is no shortcut for unlocking. If you really want to |
2264 | unlock the file prematurely, you know what to do; if you don't know |
2265 | what to do, then don't do it. |
2266 | |
2267 | All the usual warnings about file locking apply here. In particular, |
2268 | note that file locking in Perl is B<advisory>, which means that |
2269 | holding a lock will not prevent anyone else from reading, writing, or |
2270 | erasing the file; it only prevents them from getting another lock at |
2271 | the same time. Locks are analogous to green traffic lights: If you |
2272 | have a green light, that does not prevent the idiot coming the other |
2273 | way from plowing into you sideways; it merely guarantees to you that |
2274 | the idiot does not also have a green light at the same time. |
b5aed31e |
2275 | |
0b28bc9a |
2276 | =head2 C<autochomp> |
2277 | |
2278 | my $old_value = $o->autochomp(0); # disable autochomp option |
2279 | my $old_value = $o->autochomp(1); # enable autochomp option |
2280 | |
2281 | my $ac = $o->autochomp(); # recover current value |
2282 | |
2283 | See L<"autochomp">, above. |
2284 | |
6fc0ea7e |
2285 | =head2 C<defer>, C<flush>, C<discard>, and C<autodefer> |
57c7bc08 |
2286 | |
2287 | See L<"Deferred Writing">, below. |
2288 | |
6ae23f41 |
2289 | =head2 C<offset> |
2290 | |
2291 | $off = $o->offset($n); |
2292 | |
2293 | This method returns the byte offset of the start of the C<$n>th record |
2294 | in the file. If there is no such record, it returns an undefined |
2295 | value. |
2296 | |
0b28bc9a |
2297 | =head1 Tying to an already-opened filehandle |
fa408a35 |
2298 | |
2299 | If C<$fh> is a filehandle, such as is returned by C<IO::File> or one |
2300 | of the other C<IO> modules, you may use: |
2301 | |
2302 | tie @array, 'Tie::File', $fh, ...; |
2303 | |
2304 | Similarly if you opened that handle C<FH> with regular C<open> or |
2305 | C<sysopen>, you may use: |
2306 | |
2307 | tie @array, 'Tie::File', \*FH, ...; |
2308 | |
2309 | Handles that were opened write-only won't work. Handles that were |
57c7bc08 |
2310 | opened read-only will work as long as you don't try to modify the |
2311 | array. Handles must be attached to seekable sources of data---that |
27531ffb |
2312 | means no pipes or sockets. If C<Tie::File> can detect that you |
2313 | supplied a non-seekable handle, the C<tie> call will throw an |
2314 | exception. (On Unix systems, it can detect this.) |
57c7bc08 |
2315 | |
2a00e2a4 |
2316 | Note that Tie::File will only close any filehandles that it opened |
2317 | internally. If you passed it a filehandle as above, you "own" the |
2318 | filehandle, and are responsible for closing it after you have untied |
2319 | the @array. |
2320 | |
57c7bc08 |
2321 | =head1 Deferred Writing |
2322 | |
2323 | (This is an advanced feature. Skip this section on first reading.) |
2324 | |
2325 | Normally, modifying a C<Tie::File> array writes to the underlying file |
2326 | immediately. Every assignment like C<$a[3] = ...> rewrites as much of |
2327 | the file as is necessary; typically, everything from line 3 through |
2328 | the end will need to be rewritten. This is the simplest and most |
2329 | transparent behavior. Performance even for large files is reasonably |
2330 | good. |
2331 | |
2332 | However, under some circumstances, this behavior may be excessively |
2333 | slow. For example, suppose you have a million-record file, and you |
2334 | want to do: |
2335 | |
2336 | for (@FILE) { |
2337 | $_ = "> $_"; |
2338 | } |
2339 | |
2340 | The first time through the loop, you will rewrite the entire file, |
2341 | from line 0 through the end. The second time through the loop, you |
2342 | will rewrite the entire file from line 1 through the end. The third |
2343 | time through the loop, you will rewrite the entire file from line 2 to |
2344 | the end. And so on. |
2345 | |
2346 | If the performance in such cases is unacceptable, you may defer the |
2347 | actual writing, and then have it done all at once. The following loop |
2348 | will perform much better for large files: |
2349 | |
2350 | (tied @a)->defer; |
2351 | for (@a) { |
2352 | $_ = "> $_"; |
2353 | } |
2354 | (tied @a)->flush; |
2355 | |
2356 | If C<Tie::File>'s memory limit is large enough, all the writing will |
2357 | done in memory. Then, when you call C<-E<gt>flush>, the entire file |
2358 | will be rewritten in a single pass. |
2359 | |
6fc0ea7e |
2360 | (Actually, the preceding discussion is something of a fib. You don't |
2361 | need to enable deferred writing to get good performance for this |
2362 | common case, because C<Tie::File> will do it for you automatically |
2363 | unless you specifically tell it not to. See L<"autodeferring">, |
2364 | below.) |
2365 | |
57c7bc08 |
2366 | Calling C<-E<gt>flush> returns the array to immediate-write mode. If |
2367 | you wish to discard the deferred writes, you may call C<-E<gt>discard> |
2368 | instead of C<-E<gt>flush>. Note that in some cases, some of the data |
2369 | will have been written already, and it will be too late for |
6fc0ea7e |
2370 | C<-E<gt>discard> to discard all the changes. Support for |
2371 | C<-E<gt>discard> may be withdrawn in a future version of C<Tie::File>. |
57c7bc08 |
2372 | |
2373 | Deferred writes are cached in memory up to the limit specified by the |
2374 | C<dw_size> option (see above). If the deferred-write buffer is full |
2375 | and you try to write still more deferred data, the buffer will be |
2376 | flushed. All buffered data will be written immediately, the buffer |
2377 | will be emptied, and the now-empty space will be used for future |
2378 | deferred writes. |
2379 | |
2380 | If the deferred-write buffer isn't yet full, but the total size of the |
2381 | buffer and the read cache would exceed the C<memory> limit, the oldest |
27531ffb |
2382 | records will be expired from the read cache until the total size is |
57c7bc08 |
2383 | under the limit. |
2384 | |
2385 | C<push>, C<pop>, C<shift>, C<unshift>, and C<splice> cannot be |
2386 | deferred. When you perform one of these operations, any deferred data |
2387 | is written to the file and the operation is performed immediately. |
2388 | This may change in a future version. |
2389 | |
6fc0ea7e |
2390 | If you resize the array with deferred writing enabled, the file will |
2391 | be resized immediately, but deferred records will not be written. |
27531ffb |
2392 | This has a surprising consequence: C<@a = (...)> erases the file |
2393 | immediately, but the writing of the actual data is deferred. This |
2394 | might be a bug. If it is a bug, it will be fixed in a future version. |
6fc0ea7e |
2395 | |
2396 | =head2 Autodeferring |
2397 | |
2398 | C<Tie::File> tries to guess when deferred writing might be helpful, |
27531ffb |
2399 | and to turn it on and off automatically. |
2400 | |
2401 | for (@a) { |
2402 | $_ = "> $_"; |
2403 | } |
2404 | |
2405 | In this example, only the first two assignments will be done |
2406 | immediately; after this, all the changes to the file will be deferred |
2407 | up to the user-specified memory limit. |
6fc0ea7e |
2408 | |
2409 | You should usually be able to ignore this and just use the module |
2410 | without thinking about deferring. However, special applications may |
2411 | require fine control over which writes are deferred, or may require |
2412 | that all writes be immediate. To disable the autodeferment feature, |
2413 | use |
57c7bc08 |
2414 | |
2415 | (tied @o)->autodefer(0); |
2416 | |
6fc0ea7e |
2417 | or |
2418 | |
2419 | tie @array, 'Tie::File', $file, autodefer => 0; |
2420 | |
fa408a35 |
2421 | |
27531ffb |
2422 | Similarly, C<-E<gt>autodefer(1)> re-enables autodeferment, and |
2423 | C<-E<gt>autodefer()> recovers the current value of the autodefer setting. |
2424 | |
b5aed31e |
2425 | |
6ae23f41 |
2426 | =head1 CONCURRENT ACCESS TO FILES |
b5aed31e |
2427 | |
6ae23f41 |
2428 | Caching and deferred writing are inappropriate if you want the same |
bc2f65e0 |
2429 | file to be accessed simultaneously from more than one process. Other |
2430 | optimizations performed internally by this module are also |
2431 | incompatible with concurrent access. A future version of this module will |
2432 | support a C<concurrent =E<gt> 1> option that enables safe concurrent access. |
b3fe5a4c |
2433 | |
bc2f65e0 |
2434 | Previous versions of this documentation suggested using C<memory |
2435 | =E<gt> 0> for safe concurrent access. This was mistaken. Tie::File |
2436 | will not support safe concurrent access before version 0.98. |
6ae23f41 |
2437 | |
2438 | =head1 CAVEATS |
b3fe5a4c |
2439 | |
6ae23f41 |
2440 | (That's Latin for 'warnings'.) |
b3fe5a4c |
2441 | |
6ae23f41 |
2442 | =over 4 |
6fc0ea7e |
2443 | |
27531ffb |
2444 | =item * |
b5aed31e |
2445 | |
6fc0ea7e |
2446 | Reasonable effort was made to make this module efficient. Nevertheless, |
b5aed31e |
2447 | changing the size of a record in the middle of a large file will |
b3fe5a4c |
2448 | always be fairly slow, because everything after the new record must be |
2449 | moved. |
b5aed31e |
2450 | |
b3fe5a4c |
2451 | =item * |
2452 | |
2453 | The behavior of tied arrays is not precisely the same as for regular |
2454 | arrays. For example: |
b5aed31e |
2455 | |
57c7bc08 |
2456 | # This DOES print "How unusual!" |
2457 | undef $a[10]; print "How unusual!\n" if defined $a[10]; |
b3fe5a4c |
2458 | |
2459 | C<undef>-ing a C<Tie::File> array element just blanks out the |
2460 | corresponding record in the file. When you read it back again, you'll |
57c7bc08 |
2461 | get the empty string, so the supposedly-C<undef>'ed value will be |
2462 | defined. Similarly, if you have C<autochomp> disabled, then |
2463 | |
2464 | # This DOES print "How unusual!" if 'autochomp' is disabled |
27531ffb |
2465 | undef $a[10]; |
57c7bc08 |
2466 | print "How unusual!\n" if $a[10]; |
2467 | |
2468 | Because when C<autochomp> is disabled, C<$a[10]> will read back as |
2469 | C<"\n"> (or whatever the record separator string is.) |
b5aed31e |
2470 | |
6fc0ea7e |
2471 | There are other minor differences, particularly regarding C<exists> |
2472 | and C<delete>, but in general, the correspondence is extremely close. |
b3fe5a4c |
2473 | |
2474 | =item * |
2475 | |
27531ffb |
2476 | I have supposed that since this module is concerned with file I/O, |
2477 | almost all normal use of it will be heavily I/O bound. This means |
2478 | that the time to maintain complicated data structures inside the |
2479 | module will be dominated by the time to actually perform the I/O. |
2480 | When there was an opportunity to spend CPU time to avoid doing I/O, I |
6ae23f41 |
2481 | usually tried to take it. |
b5aed31e |
2482 | |
57c7bc08 |
2483 | =item * |
6fc0ea7e |
2484 | |
57c7bc08 |
2485 | You might be tempted to think that deferred writing is like |
2486 | transactions, with C<flush> as C<commit> and C<discard> as |
6fc0ea7e |
2487 | C<rollback>, but it isn't, so don't. |
57c7bc08 |
2488 | |
6ae23f41 |
2489 | =item * |
2490 | |
2491 | There is a large memory overhead for each record offset and for each |
2492 | cache entry: about 310 bytes per cached data record, and about 21 bytes per offset table entry. |
2493 | |
2494 | The per-record overhead will limit the maximum number of records you |
2495 | can access per file. Note that I<accessing> the length of the array |
2496 | via C<$x = scalar @tied_file> accesses B<all> records and stores their |
2497 | offsets. The same for C<foreach (@tied_file)>, even if you exit the |
2498 | loop early. |
2499 | |
b3fe5a4c |
2500 | =back |
51efdd02 |
2501 | |
57c7bc08 |
2502 | =head1 SUBCLASSING |
2503 | |
2504 | This version promises absolutely nothing about the internals, which |
2505 | may change without notice. A future version of the module will have a |
2506 | well-defined and stable subclassing API. |
2507 | |
b3fe5a4c |
2508 | =head1 WHAT ABOUT C<DB_File>? |
51efdd02 |
2509 | |
27531ffb |
2510 | People sometimes point out that L<DB_File> will do something similar, |
2511 | and ask why C<Tie::File> module is necessary. |
b3fe5a4c |
2512 | |
27531ffb |
2513 | There are a number of reasons that you might prefer C<Tie::File>. |
2514 | A list is available at C<http://perl.plover.com/TieFile/why-not-DB_File>. |
b5aed31e |
2515 | |
2516 | =head1 AUTHOR |
2517 | |
2518 | Mark Jason Dominus |
2519 | |
2520 | To contact the author, send email to: C<mjd-perl-tiefile+@plover.com> |
2521 | |
2522 | To receive an announcement whenever a new version of this module is |
2523 | released, send a blank email message to |
2524 | C<mjd-perl-tiefile-subscribe@plover.com>. |
2525 | |
57c7bc08 |
2526 | The most recent version of this module, including documentation and |
2527 | any news of importance, will be available at |
2528 | |
2529 | http://perl.plover.com/TieFile/ |
2530 | |
2531 | |
b5aed31e |
2532 | =head1 LICENSE |
2533 | |
bc2f65e0 |
2534 | C<Tie::File> version 0.97 is copyright (C) 2003 Mark Jason Dominus. |
7b6b3db1 |
2535 | |
2536 | This library is free software; you may redistribute it and/or modify |
2537 | it under the same terms as Perl itself. |
b5aed31e |
2538 | |
57c7bc08 |
2539 | These terms are your choice of any of (1) the Perl Artistic Licence, |
2540 | or (2) version 2 of the GNU General Public License as published by the |
7b6b3db1 |
2541 | Free Software Foundation, or (3) any later version of the GNU General |
2542 | Public License. |
b5aed31e |
2543 | |
7b6b3db1 |
2544 | This library is distributed in the hope that it will be useful, |
b5aed31e |
2545 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
2546 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
2547 | GNU General Public License for more details. |
2548 | |
2549 | You should have received a copy of the GNU General Public License |
7b6b3db1 |
2550 | along with this library program; it should be in the file C<COPYING>. |
2551 | If not, write to the Free Software Foundation, Inc., 59 Temple Place, |
2552 | Suite 330, Boston, MA 02111 USA |
b5aed31e |
2553 | |
2554 | For licensing inquiries, contact the author at: |
2555 | |
2556 | Mark Jason Dominus |
2557 | 255 S. Warnock St. |
2558 | Philadelphia, PA 19107 |
2559 | |
2560 | =head1 WARRANTY |
2561 | |
bc2f65e0 |
2562 | C<Tie::File> version 0.97 comes with ABSOLUTELY NO WARRANTY. |
b5aed31e |
2563 | For details, see the license. |
2564 | |
fa408a35 |
2565 | =head1 THANKS |
2566 | |
2567 | Gigantic thanks to Jarkko Hietaniemi, for agreeing to put this in the |
2568 | core when I hadn't written it yet, and for generally being helpful, |
2569 | supportive, and competent. (Usually the rule is "choose any one.") |
2570 | Also big thanks to Abhijit Menon-Sen for all of the same things. |
2571 | |
57c7bc08 |
2572 | Special thanks to Craig Berry and Peter Prymmer (for VMS portability |
2573 | help), Randy Kobes (for Win32 portability help), Clinton Pierce and |
2574 | Autrijus Tang (for heroic eleventh-hour Win32 testing above and beyond |
6fc0ea7e |
2575 | the call of duty), Michael G Schwern (for testing advice), and the |
2576 | rest of the CPAN testers (for testing generally). |
b5aed31e |
2577 | |
6ae23f41 |
2578 | Special thanks to Tels for suggesting several speed and memory |
2579 | optimizations. |
2580 | |
57c7bc08 |
2581 | Additional thanks to: |
b3fe5a4c |
2582 | Edward Avis / |
6ae23f41 |
2583 | Mattia Barbon / |
29aee836 |
2584 | Tom Christiansen / |
fa408a35 |
2585 | Gerrit Haase / |
29aee836 |
2586 | Gurusamy Sarathy / |
6ae23f41 |
2587 | Jarkko Hietaniemi (again) / |
b3fe5a4c |
2588 | Nikola Knezevic / |
6ae23f41 |
2589 | John Kominetz / |
836d9961 |
2590 | Nick Ing-Simmons / |
fa408a35 |
2591 | Tassilo von Parseval / |
2592 | H. Dieter Pearcey / |
b3fe5a4c |
2593 | Slaven Rezic / |
6ae23f41 |
2594 | Eric Roode / |
6fc0ea7e |
2595 | Peter Scott / |
fa408a35 |
2596 | Peter Somu / |
57c7bc08 |
2597 | Autrijus Tang (again) / |
6ae23f41 |
2598 | Tels (again) / |
2599 | Juerd Waalboer |
7b6b3db1 |
2600 | |
fa408a35 |
2601 | =head1 TODO |
2602 | |
b5aed31e |
2603 | More tests. (Stuff I didn't think of yet.) |
2604 | |
b5aed31e |
2605 | Paragraph mode? |
2606 | |
6fc0ea7e |
2607 | Fixed-length mode. Leave-blanks mode. |
b5aed31e |
2608 | |
fa408a35 |
2609 | Maybe an autolocking mode? |
2610 | |
6ae23f41 |
2611 | For many common uses of the module, the read cache is a liability. |
2612 | For example, a program that inserts a single record, or that scans the |
2613 | file once, will have a cache hit rate of zero. This suggests a major |
2614 | optimization: The cache should be initially disabled. Here's a hybrid |
2615 | approach: Initially, the cache is disabled, but the cache code |
2616 | maintains statistics about how high the hit rate would be *if* it were |
2617 | enabled. When it sees the hit rate get high enough, it enables |
2618 | itself. The STAT comments in this code are the beginning of an |
2619 | implementation of this. |
2620 | |
6fc0ea7e |
2621 | Record locking with fcntl()? Then the module might support an undo |
2622 | log and get real transactions. What a tour de force that would be. |
b3fe5a4c |
2623 | |
6ae23f41 |
2624 | Keeping track of the highest cached record. This would allow reads-in-a-row |
2625 | to skip the cache lookup faster (if reading from 1..N with empty cache at |
2626 | start, the last cached value will be always N-1). |
2627 | |
27531ffb |
2628 | More tests. |
b3fe5a4c |
2629 | |
b5aed31e |
2630 | =cut |
2631 | |