3 perldebguts - Guts of Perl debugging
7 This is not the perldebug(1) manpage, which tells you how to use
8 the debugger. This manpage describes low-level details ranging
9 between difficult and impossible for anyone who isn't incredibly
10 intimate with Perl's guts to understand. Caveat lector.
12 =head1 Debugger Internals
14 Perl has special debugging hooks at compile-time and run-time used
15 to create debugging environments. These hooks are not to be confused
16 with the I<perl -Dxxx> command described in L<perlrun>, which is
17 usable only if a special Perl is built per the instructions in the
18 F<INSTALL> podpage in the Perl source tree.
20 For example, whenever you call Perl's built-in C<caller> function
21 from the package DB, the arguments that the corresponding stack
22 frame was called with are copied to the C<@DB::args> array. The
23 general mechanism is enabled by calling Perl with the B<-d> switch, the
24 following additional features are enabled (cf. L<perlvar/$^P>):
30 Perl inserts the contents of C<$ENV{PERL5DB}> (or C<BEGIN {require
31 'perl5db.pl'}> if not present) before the first line of your program.
35 Each array C<@{"_<$filename"}> holds the lines of $filename for a
36 file compiled by Perl. The same for C<eval>ed strings that contain
37 subroutines, or which are currently being executed. The $filename
38 for C<eval>ed strings looks like C<(eval 34)>. Code assertions
39 in regexes look like C<(re_eval 19)>.
41 Values in this array are magical in numeric context: they compare
42 equal to zero only if the line is not breakable.
46 Each hash C<%{"_<$filename"}> contains breakpoints and actions keyed
47 by line number. Individual entries (as opposed to the whole hash)
48 are settable. Perl only cares about Boolean true here, although
49 the values used by F<perl5db.pl> have the form
50 C<"$break_condition\0$action">.
52 The same holds for evaluated strings that contain subroutines, or
53 which are currently being executed. The $filename for C<eval>ed strings
54 looks like C<(eval 34)> or C<(re_eval 19)>.
58 Each scalar C<${"_<$filename"}> contains C<"_<$filename">. This is
59 also the case for evaluated strings that contain subroutines, or
60 which are currently being executed. The $filename for C<eval>ed
61 strings looks like C<(eval 34)> or C<(re_eval 19)>.
65 After each C<require>d file is compiled, but before it is executed,
66 C<DB::postponed(*{"_<$filename"})> is called if the subroutine
67 C<DB::postponed> exists. Here, the $filename is the expanded name of
68 the C<require>d file, as found in the values of %INC.
72 After each subroutine C<subname> is compiled, the existence of
73 C<$DB::postponed{subname}> is checked. If this key exists,
74 C<DB::postponed(subname)> is called if the C<DB::postponed> subroutine
79 A hash C<%DB::sub> is maintained, whose keys are subroutine names
80 and whose values have the form C<filename:startline-endline>.
81 C<filename> has the form C<(eval 34)> for subroutines defined inside
82 C<eval>s, or C<(re_eval 19)> for those within regex code assertions.
86 When the execution of your program reaches a point that can hold a
87 breakpoint, the C<DB::DB()> subroutine is called any of the variables
88 $DB::trace, $DB::single, or $DB::signal is true. These variables
89 are not C<local>izable. This feature is disabled when executing
90 inside C<DB::DB()>, including functions called from it
91 unless C<< $^D & (1<<30) >> is true.
95 When execution of the program reaches a subroutine call, a call to
96 C<&DB::sub>(I<args>) is made instead, with C<$DB::sub> holding the
97 name of the called subroutine. This doesn't happen if the subroutine
98 was compiled in the C<DB> package.)
102 Note that if C<&DB::sub> needs external data for it to work, no
103 subroutine call is possible until this is done. For the standard
104 debugger, the C<$DB::deep> variable (how many levels of recursion
105 deep into the debugger you can go before a mandatory break) gives
106 an example of such a dependency.
108 =head2 Writing Your Own Debugger
110 The minimal working debugger consists of one line
114 which you could even fit into the C<PERL5DB> environment
117 $ PERL5DB="sub DB::DB {}" perl -d your-script
119 although it doesn't do anything that tells you it's working...
120 Another brief debugger, slightly more useful, could be created
123 sub DB::DB {print ++$i; scalar <STDIN>}
125 This debugger would print the sequential number of encountered
126 statement, and would wait for you to hit a newline before continuing.
128 The following debugger is quite functional:
133 sub sub {print ++$i, " $sub\n"; &$sub}
136 It prints the sequential number of subroutine call and the name of the
137 called subroutine. Note that C<&DB::sub> should be compiled into the
140 At the start, the debugger reads your rc file (F<./.perldb> or
141 F<~/.perldb> under Unix), which can set important options. This file may
142 define a subroutine C<&afterinit> to be executed after the debugger is
145 After the rc file is read, the debugger reads the PERLDB_OPTS
146 environment variable and parses this as the remainder of a C<O ...>
147 line as one might enter at the debugger prompt.
149 The debugger also maintains magical internal variables, such as
150 C<@DB::dbline>, C<%DB::dbline>, which are aliases for
151 C<@{"::_<current_file"}> C<%{"::_<current_file"}>. Here C<current_file>
152 is the currently selected file, either explicitly chosen with the
153 debugger's C<f> command, or implicitly by flow of execution.
155 Some functions are provided to simplify customization. See
156 L<perldebug/"Options"> for description of options parsed by
157 C<DB::parse_options(string)>. The function C<DB::dump_trace(skip[,
158 count])> skips the specified number of frames and returns a list
159 containing information about the calling frames (all of them, if
160 C<count> is missing). Each entry is reference to a hash with
161 keys C<context> (either C<.>, C<$>, or C<@>), C<sub> (subroutine
162 name, or info about C<eval>), C<args> (C<undef> or a reference to
163 an array), C<file>, and C<line>.
165 The function C<DB::print_trace(FH, skip[, count[, short]])> prints
166 formatted info about caller frames. The last two functions may be
167 convenient as arguments to C<< < >>, C<< << >> commands.
169 Note that any variables and functions that are not documented in
170 this manpages (or in L<perldebug>) are considered for internal
171 use only, and as such are subject to change without notice.
173 =head1 Frame Listing Output Examples
175 The C<frame> option can be used to control the output of frame
176 information. For example, contrast this expression trace:
179 Stack dump during die enabled outside of evals.
181 Loading DB routines from perl5db.pl patch level 0.94
182 Emacs support available.
184 Enter h or `h h' for help.
191 DB<3> t print foo() * bar()
192 main::((eval 172):3): print foo() + bar();
193 main::foo((eval 168):2):
194 main::bar((eval 170):2):
197 with this one, once the C<O>ption C<frame=2> has been set:
201 DB<5> t print foo() * bar()
211 By way of demonstration, we present below a laborious listing
212 resulting from setting your C<PERLDB_OPTS> environment variable to
213 the value C<f=n N>, and running I<perl -d -V> from the command line.
214 Examples use various values of C<n> are shown to give you a feel
215 for the difference between settings. Long those it may be, this
216 is not a complete listing, but only excerpts.
223 entering Config::BEGIN
224 Package lib/Exporter.pm.
226 Package lib/Config.pm.
227 entering Config::TIEHASH
228 entering Exporter::import
229 entering Exporter::export
230 entering Config::myconfig
231 entering Config::FETCH
232 entering Config::FETCH
233 entering Config::FETCH
234 entering Config::FETCH
239 entering Config::BEGIN
240 Package lib/Exporter.pm.
243 Package lib/Config.pm.
244 entering Config::TIEHASH
245 exited Config::TIEHASH
246 entering Exporter::import
247 entering Exporter::export
248 exited Exporter::export
249 exited Exporter::import
251 entering Config::myconfig
252 entering Config::FETCH
254 entering Config::FETCH
256 entering Config::FETCH
260 in $=main::BEGIN() from /dev/null:0
261 in $=Config::BEGIN() from lib/Config.pm:2
262 Package lib/Exporter.pm.
264 Package lib/Config.pm.
265 in $=Config::TIEHASH('Config') from lib/Config.pm:644
266 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
267 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from li
268 in @=Config::myconfig() from /dev/null:0
269 in $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
270 in $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
271 in $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
272 in $=Config::FETCH(ref(Config), 'PERL_SUBVERSION') from lib/Config.pm:574
273 in $=Config::FETCH(ref(Config), 'osname') from lib/Config.pm:574
274 in $=Config::FETCH(ref(Config), 'osvers') from lib/Config.pm:574
278 in $=main::BEGIN() from /dev/null:0
279 in $=Config::BEGIN() from lib/Config.pm:2
280 Package lib/Exporter.pm.
282 out $=Config::BEGIN() from lib/Config.pm:0
283 Package lib/Config.pm.
284 in $=Config::TIEHASH('Config') from lib/Config.pm:644
285 out $=Config::TIEHASH('Config') from lib/Config.pm:644
286 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
287 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/
288 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/
289 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
290 out $=main::BEGIN() from /dev/null:0
291 in @=Config::myconfig() from /dev/null:0
292 in $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
293 out $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
294 in $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
295 out $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
296 in $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
297 out $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
298 in $=Config::FETCH(ref(Config), 'PERL_SUBVERSION') from lib/Config.pm:574
302 in $=main::BEGIN() from /dev/null:0
303 in $=Config::BEGIN() from lib/Config.pm:2
304 Package lib/Exporter.pm.
306 out $=Config::BEGIN() from lib/Config.pm:0
307 Package lib/Config.pm.
308 in $=Config::TIEHASH('Config') from lib/Config.pm:644
309 out $=Config::TIEHASH('Config') from lib/Config.pm:644
310 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
311 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/E
312 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/E
313 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
314 out $=main::BEGIN() from /dev/null:0
315 in @=Config::myconfig() from /dev/null:0
316 in $=Config::FETCH('Config=HASH(0x1aa444)', 'package') from lib/Config.pm:574
317 out $=Config::FETCH('Config=HASH(0x1aa444)', 'package') from lib/Config.pm:574
318 in $=Config::FETCH('Config=HASH(0x1aa444)', 'baserev') from lib/Config.pm:574
319 out $=Config::FETCH('Config=HASH(0x1aa444)', 'baserev') from lib/Config.pm:574
323 in $=CODE(0x15eca4)() from /dev/null:0
324 in $=CODE(0x182528)() from lib/Config.pm:2
325 Package lib/Exporter.pm.
326 out $=CODE(0x182528)() from lib/Config.pm:0
327 scalar context return from CODE(0x182528): undef
328 Package lib/Config.pm.
329 in $=Config::TIEHASH('Config') from lib/Config.pm:628
330 out $=Config::TIEHASH('Config') from lib/Config.pm:628
331 scalar context return from Config::TIEHASH: empty hash
332 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
333 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/Exporter.pm:171
334 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/Exporter.pm:171
335 scalar context return from Exporter::export: ''
336 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
337 scalar context return from Exporter::import: ''
341 In all cases shown above, the line indentation shows the call tree.
342 If bit 2 of C<frame> is set, a line is printed on exit from a
343 subroutine as well. If bit 4 is set, the arguments are printed
344 along with the caller info. If bit 8 is set, the arguments are
345 printed even if they are tied or references. If bit 16 is set, the
346 return value is printed, too.
348 When a package is compiled, a line like this
352 is printed with proper indentation.
354 =head1 Debugging regular expressions
356 There are two ways to enable debugging output for regular expressions.
358 If your perl is compiled with C<-DDEBUGGING>, you may use the
359 B<-Dr> flag on the command line.
361 Otherwise, one can C<use re 'debug'>, which has effects at
362 compile time and run time. It is not lexically scoped.
364 =head2 Compile-time output
366 The debugging output at compile time looks like this:
368 Compiling REx `[bc]d(ef*g)+h[ij]k$'
369 size 45 Got 364 bytes for offset annotations.
375 14: CURLYX[0] {1,32767}(28)
389 anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating)
390 stclass `ANYOF[bc]' minlen 7
392 1[4] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 5[1]
393 0[0] 12[1] 0[0] 6[1] 0[0] 7[1] 0[0] 9[1] 8[1] 0[0] 10[1] 0[0]
394 11[1] 0[0] 12[0] 12[0] 13[1] 0[0] 14[4] 0[0] 0[0] 0[0] 0[0]
395 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 18[1] 0[0] 19[1] 20[0]
396 Omitting $` $& $' support.
398 The first line shows the pre-compiled form of the regex. The second
399 shows the size of the compiled form (in arbitrary units, usually
400 4-byte words) and the total number of bytes allocated for the
401 offset/length table, usually 4+C<size>*8. The next line shows the
402 label I<id> of the first node that does a match.
406 anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating)
407 stclass `ANYOF[bc]' minlen 7
409 line (split into two lines above) contains optimizer
410 information. In the example shown, the optimizer found that the match
411 should contain a substring C<de> at offset 1, plus substring C<gh>
412 at some offset between 3 and infinity. Moreover, when checking for
413 these substrings (to abandon impossible matches quickly), Perl will check
414 for the substring C<gh> before checking for the substring C<de>. The
415 optimizer may also use the knowledge that the match starts (at the
416 C<first> I<id>) with a character class, and no string
417 shorter than 7 characters can possibly match.
419 The fields of interest which may appear in this line are
423 =item C<anchored> I<STRING> C<at> I<POS>
425 =item C<floating> I<STRING> C<at> I<POS1..POS2>
429 =item C<matching floating/anchored>
431 Which substring to check first.
435 The minimal length of the match.
437 =item C<stclass> I<TYPE>
439 Type of first matching node.
443 Don't scan for the found substrings.
447 Means that the optimizer information is all that the regular
448 expression contains, and thus one does not need to enter the regex engine at
453 Set if the pattern contains C<\G>.
457 Set if the pattern starts with a repeated char (as in C<x+y>).
461 Set if the pattern starts with C<.*>.
465 Set if the pattern contain eval-groups, such as C<(?{ code })> and
468 =item C<anchored(TYPE)>
470 If the pattern may match only at a handful of places, (with C<TYPE>
471 being C<BOL>, C<MBOL>, or C<GPOS>. See the table below.
475 If a substring is known to match at end-of-line only, it may be
476 followed by C<$>, as in C<floating `k'$>.
478 The optimizer-specific information is used to avoid entering (a slow) regex
479 engine on strings that will not definitely match. If the C<isall> flag
480 is set, a call to the regex engine may be avoided even when the optimizer
481 found an appropriate place for the match.
483 Above the optimizer section is the list of I<nodes> of the compiled
484 form of the regex. Each line has format
486 C< >I<id>: I<TYPE> I<OPTIONAL-INFO> (I<next-id>)
488 =head2 Types of nodes
490 Here are the possible types, with short descriptions:
492 # TYPE arg-description [num-args] [longjump-len] DESCRIPTION
495 END no End of program.
496 SUCCEED no Return from a subroutine, basically.
499 BOL no Match "" at beginning of line.
500 MBOL no Same, assuming multiline.
501 SBOL no Same, assuming singleline.
502 EOS no Match "" at end of string.
503 EOL no Match "" at end of line.
504 MEOL no Same, assuming multiline.
505 SEOL no Same, assuming singleline.
506 BOUND no Match "" at any word boundary
507 BOUNDL no Match "" at any word boundary
508 NBOUND no Match "" at any word non-boundary
509 NBOUNDL no Match "" at any word non-boundary
510 GPOS no Matches where last m//g left off.
512 # [Special] alternatives
513 ANY no Match any one character (except newline).
514 SANY no Match any one character.
515 ANYOF sv Match character in (or not in) this class.
516 ALNUM no Match any alphanumeric character
517 ALNUML no Match any alphanumeric char in locale
518 NALNUM no Match any non-alphanumeric character
519 NALNUML no Match any non-alphanumeric char in locale
520 SPACE no Match any whitespace character
521 SPACEL no Match any whitespace char in locale
522 NSPACE no Match any non-whitespace character
523 NSPACEL no Match any non-whitespace char in locale
524 DIGIT no Match any numeric character
525 NDIGIT no Match any non-numeric character
527 # BRANCH The set of branches constituting a single choice are hooked
528 # together with their "next" pointers, since precedence prevents
529 # anything being concatenated to any individual branch. The
530 # "next" pointer of the last BRANCH in a choice points to the
531 # thing following the whole choice. This is also where the
532 # final "next" pointer of each individual branch points; each
533 # branch starts with the operand node of a BRANCH node.
535 BRANCH node Match this alternative, or the next...
537 # BACK Normal "next" pointers all implicitly point forward; BACK
538 # exists to make loop structures possible.
540 BACK no Match "", "next" ptr points backward.
543 EXACT sv Match this string (preceded by length).
544 EXACTF sv Match this string, folded (prec. by length).
545 EXACTFL sv Match this string, folded in locale (w/len).
548 NOTHING no Match empty string.
549 # A variant of above which delimits a group, thus stops optimizations
550 TAIL no Match empty string. Can jump here from outside.
552 # STAR,PLUS '?', and complex '*' and '+', are implemented as circular
553 # BRANCH structures using BACK. Simple cases (one character
554 # per match) are implemented with STAR and PLUS for speed
555 # and to minimize recursive plunges.
557 STAR node Match this (simple) thing 0 or more times.
558 PLUS node Match this (simple) thing 1 or more times.
560 CURLY sv 2 Match this simple thing {n,m} times.
561 CURLYN no 2 Match next-after-this simple thing
562 # {n,m} times, set parens.
563 CURLYM no 2 Match this medium-complex thing {n,m} times.
564 CURLYX sv 2 Match this complex thing {n,m} times.
566 # This terminator creates a loop structure for CURLYX
567 WHILEM no Do curly processing and see if rest matches.
569 # OPEN,CLOSE,GROUPP ...are numbered at compile time.
570 OPEN num 1 Mark this point in input as start of #n.
571 CLOSE num 1 Analogous to OPEN.
573 REF num 1 Match some already matched string
574 REFF num 1 Match already matched string, folded
575 REFFL num 1 Match already matched string, folded in loc.
577 # grouping assertions
578 IFMATCH off 1 2 Succeeds if the following matches.
579 UNLESSM off 1 2 Fails if the following matches.
580 SUSPEND off 1 1 "Independent" sub-regex.
581 IFTHEN off 1 1 Switch, should be preceded by switcher .
582 GROUPP num 1 Whether the group matched.
584 # Support for long regex
585 LONGJMP off 1 1 Jump far away.
586 BRANCHJ off 1 1 BRANCH with long offset.
589 EVAL evl 1 Execute some Perl code.
592 MINMOD no Next operator is not greedy.
593 LOGICAL no Next opcode should set the flag only.
595 # This is not used yet
596 RENUM off 1 1 Group with independently numbered parens.
598 # This is not really a node, but an optimized away piece of a "long" node.
599 # To simplify debugging output, we mark it as if it were a node
600 OPTIMIZED off Placeholder for dump.
602 =for unprinted-credits
603 Next section M-J. Dominus (mjd-perl-patch+@plover.com) 20010421
605 Following the optimizer information is a dump of the offset/length
606 table, here split across several lines:
609 1[4] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 5[1]
610 0[0] 12[1] 0[0] 6[1] 0[0] 7[1] 0[0] 9[1] 8[1] 0[0] 10[1] 0[0]
611 11[1] 0[0] 12[0] 12[0] 13[1] 0[0] 14[4] 0[0] 0[0] 0[0] 0[0]
612 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 18[1] 0[0] 19[1] 20[0]
614 The first line here indicates that the offset/length table contains 45
615 entries. Each entry is a pair of integers, denoted by C<offset[length]>.
616 Entries are numbered starting with 1, so entry #1 here is C<1[4]> and
617 entry #12 is C<5[1]>. C<1[4]> indicates that the node labeled C<1:>
618 (the C<1: ANYOF[bc]>) begins at character position 1 in the
619 pre-compiled form of the regex, and has a length of 4 characters.
620 C<5[1]> in position 12
621 indicates that the node labeled C<12:>
622 (the C<< 12: EXACT <d> >>) begins at character position 5 in the
623 pre-compiled form of the regex, and has a length of 1 character.
624 C<12[1]> in position 14
625 indicates that the node labeled C<14:>
626 (the C<< 14: CURLYX[0] {1,32767} >>) begins at character position 12 in the
627 pre-compiled form of the regex, and has a length of 1 character---that
628 is, it corresponds to the C<+> symbol in the precompiled regex.
630 C<0[0]> items indicate that there is no corresponding node.
632 =head2 Run-time output
634 First of all, when doing a match, one may get no run-time output even
635 if debugging is enabled. This means that the regex engine was never
636 entered and that all of the job was therefore done by the optimizer.
638 If the regex engine was entered, the output may look like this:
640 Matching `[bc]d(ef*g)+h[ij]k$' against `abcdefg__gh__'
641 Setting an EVAL scope, savestack=3
642 2 <ab> <cdefg__gh_> | 1: ANYOF
643 3 <abc> <defg__gh_> | 11: EXACT <d>
644 4 <abcd> <efg__gh_> | 13: CURLYX {1,32767}
645 4 <abcd> <efg__gh_> | 26: WHILEM
646 0 out of 1..32767 cc=effff31c
647 4 <abcd> <efg__gh_> | 15: OPEN1
648 4 <abcd> <efg__gh_> | 17: EXACT <e>
649 5 <abcde> <fg__gh_> | 19: STAR
650 EXACT <f> can match 1 times out of 32767...
651 Setting an EVAL scope, savestack=3
652 6 <bcdef> <g__gh__> | 22: EXACT <g>
653 7 <bcdefg> <__gh__> | 24: CLOSE1
654 7 <bcdefg> <__gh__> | 26: WHILEM
655 1 out of 1..32767 cc=effff31c
656 Setting an EVAL scope, savestack=12
657 7 <bcdefg> <__gh__> | 15: OPEN1
658 7 <bcdefg> <__gh__> | 17: EXACT <e>
659 restoring \1 to 4(4)..7
660 failed, try continuation...
661 7 <bcdefg> <__gh__> | 27: NOTHING
662 7 <bcdefg> <__gh__> | 28: EXACT <h>
666 The most significant information in the output is about the particular I<node>
667 of the compiled regex that is currently being tested against the target string.
668 The format of these lines is
670 C< >I<STRING-OFFSET> <I<PRE-STRING>> <I<POST-STRING>> |I<ID>: I<TYPE>
672 The I<TYPE> info is indented with respect to the backtracking level.
673 Other incidental information appears interspersed within.
675 =head1 Debugging Perl memory usage
677 Perl is a profligate wastrel when it comes to memory use. There
678 is a saying that to estimate memory usage of Perl, assume a reasonable
679 algorithm for memory allocation, multiply that estimate by 10, and
680 while you still may miss the mark, at least you won't be quite so
681 astonished. This is not absolutely true, but may provide a good
682 grasp of what happens.
684 Assume that an integer cannot take less than 20 bytes of memory, a
685 float cannot take less than 24 bytes, a string cannot take less
686 than 32 bytes (all these examples assume 32-bit architectures, the
687 result are quite a bit worse on 64-bit architectures). If a variable
688 is accessed in two of three different ways (which require an integer,
689 a float, or a string), the memory footprint may increase yet another
690 20 bytes. A sloppy malloc(3) implementation can inflate these
691 numbers dramatically.
693 On the opposite end of the scale, a declaration like
697 may take up to 500 bytes of memory, depending on which release of Perl
700 Anecdotal estimates of source-to-compiled code bloat suggest an
701 eightfold increase. This means that the compiled form of reasonable
702 (normally commented, properly indented etc.) code will take
703 about eight times more space in memory than the code took
706 There are two Perl-specific ways to analyze memory usage:
707 $ENV{PERL_DEBUG_MSTATS} and B<-DL> command-line switch. The first
708 is available only if Perl is compiled with Perl's malloc(); the
709 second only if Perl was built with C<-DDEBUGGING>. See the
710 instructions for how to do this in the F<INSTALL> podpage at
711 the top level of the Perl source tree.
713 =head2 Using C<$ENV{PERL_DEBUG_MSTATS}>
715 If your perl is using Perl's malloc() and was compiled with the
716 necessary switches (this is the default), then it will print memory
717 usage statistics after compiling your code when C<< $ENV{PERL_DEBUG_MSTATS}
718 > 1 >>, and before termination of the program when C<<
719 $ENV{PERL_DEBUG_MSTATS} >= 1 >>. The report format is similar to
720 the following example:
722 $ PERL_DEBUG_MSTATS=2 perl -e "require Carp"
723 Memory allocation statistics after compilation: (buckets 4(4)..8188(8192)
724 14216 free: 130 117 28 7 9 0 2 2 1 0 0
726 60924 used: 125 137 161 55 7 8 6 16 2 0 1
728 Total sbrk(): 77824/21:119. Odd ends: pad+heads+chain+tail: 0+636+0+2048.
729 Memory allocation statistics after execution: (buckets 4(4)..8188(8192)
730 30888 free: 245 78 85 13 6 2 1 3 2 0 1
732 175816 used: 265 176 1112 111 26 22 11 27 2 1 1
734 Total sbrk(): 215040/47:145. Odd ends: pad+heads+chain+tail: 0+2192+0+6144.
736 It is possible to ask for such a statistic at arbitrary points in
737 your execution using the mstat() function out of the standard
740 Here is some explanation of that format:
744 =item C<buckets SMALLEST(APPROX)..GREATEST(APPROX)>
746 Perl's malloc() uses bucketed allocations. Every request is rounded
747 up to the closest bucket size available, and a bucket is taken from
748 the pool of buckets of that size.
750 The line above describes the limits of buckets currently in use.
751 Each bucket has two sizes: memory footprint and the maximal size
752 of user data that can fit into this bucket. Suppose in the above
753 example that the smallest bucket were size 4. The biggest bucket
754 would have usable size 8188, and the memory footprint would be 8192.
756 In a Perl built for debugging, some buckets may have negative usable
757 size. This means that these buckets cannot (and will not) be used.
758 For larger buckets, the memory footprint may be one page greater
759 than a power of 2. If so, case the corresponding power of two is
760 printed in the C<APPROX> field above.
764 The 1 or 2 rows of numbers following that correspond to the number
765 of buckets of each size between C<SMALLEST> and C<GREATEST>. In
766 the first row, the sizes (memory footprints) of buckets are powers
767 of two--or possibly one page greater. In the second row, if present,
768 the memory footprints of the buckets are between the memory footprints
769 of two buckets "above".
771 For example, suppose under the previous example, the memory footprints
774 free: 8 16 32 64 128 256 512 1024 2048 4096 8192
777 With non-C<DEBUGGING> perl, the buckets starting from C<128> have
778 a 4-byte overhead, and thus an 8192-long bucket may take up to
779 8188-byte allocations.
781 =item C<Total sbrk(): SBRKed/SBRKs:CONTINUOUS>
783 The first two fields give the total amount of memory perl sbrk(2)ed
784 (ess-broken? :-) and number of sbrk(2)s used. The third number is
785 what perl thinks about continuity of returned chunks. So long as
786 this number is positive, malloc() will assume that it is probable
787 that sbrk(2) will provide continuous memory.
789 Memory allocated by external libraries is not counted.
793 The amount of sbrk(2)ed memory needed to keep buckets aligned.
797 Although memory overhead of bigger buckets is kept inside the bucket, for
798 smaller buckets, it is kept in separate areas. This field gives the
799 total size of these areas.
803 malloc() may want to subdivide a bigger bucket into smaller buckets.
804 If only a part of the deceased bucket is left unsubdivided, the rest
805 is kept as an element of a linked list. This field gives the total
806 size of these chunks.
810 To minimize the number of sbrk(2)s, malloc() asks for more memory. This
811 field gives the size of the yet unused part, which is sbrk(2)ed, but
816 =head2 Example of using B<-DL> switch
818 Below we show how to analyse memory usage by
820 do 'lib/auto/POSIX/autosplit.ix';
822 The file in question contains a header and 146 lines similar to
826 B<WARNING>: The discussion below supposes 32-bit architecture. In
827 newer releases of Perl, memory usage of the constructs discussed
828 here is greatly improved, but the story discussed below is a real-life
829 story. This story is mercilessly terse, and assumes rather more than cursory
830 knowledge of Perl internals. Type space to continue, `q' to quit.
831 (Actually, you just want to skip to the next section.)
833 Here is the itemized list of Perl allocations performed during parsing
836 !!! "after" at test.pl line 3.
837 Id subtot 4 8 12 16 20 24 28 32 36 40 48 56 64 72 80 80+
838 0 02 13752 . . . . 294 . . . . . . . . . . 4
839 0 54 5545 . . 8 124 16 . . . 1 1 . . . . . 3
840 5 05 32 . . . . . . . 1 . . . . . . . .
841 6 02 7152 . . . . . . . . . . 149 . . . . .
842 7 02 3600 . . . . . 150 . . . . . . . . . .
843 7 03 64 . -1 . 1 . . 2 . . . . . . . . .
844 7 04 7056 . . . . . . . . . . . . . . . 7
845 7 17 38404 . . . . . . . 1 . . 442 149 . . 147 .
846 9 03 2078 17 249 32 . . . . 2 . . . . . . . .
849 To see this list, insert two C<warn('!...')> statements around the call:
852 do 'lib/auto/POSIX/autosplit.ix';
855 and run it with Perl's B<-DL> option. The first warn() will print
856 memory allocation info before parsing the file and will memorize
857 the statistics at this point (we ignore what it prints). The second
858 warn() prints increments with respect to these memorized data. This
859 is the printout shown above.
861 Different I<Id>s on the left correspond to different subsystems of
862 the perl interpreter. They are just the first argument given to
863 the perl memory allocation API named New(). To find what C<9 03>
864 means, just B<grep> the perl source for C<903>. You'll find it in
865 F<util.c>, function savepvn(). (I know, you wonder why we told you
866 to B<grep> and then gave away the answer. That's because grepping
867 the source is good for the soul.) This function is used to store
868 a copy of an existing chunk of memory. Using a C debugger, one can
869 see that the function was called either directly from gv_init() or
870 via sv_magic(), and that gv_init() is called from gv_fetchpv()--which
871 was itself called from newSUB(). Please stop to catch your breath now.
873 B<NOTE>: To reach this point in the debugger and skip the calls to
874 savepvn() during the compilation of the main program, you should
876 in Perl_warn(), continue until this point is reached, and I<then> set
877 a C breakpoint in Perl_savepvn(). Note that you may need to skip a
878 handful of Perl_savepvn() calls that do not correspond to mass production
879 of CVs (there are more C<903> allocations than 146 similar lines of
880 F<lib/auto/POSIX/autosplit.ix>). Note also that C<Perl_> prefixes are
881 added by macroization code in perl header files to avoid conflicts
882 with external libraries.
884 Anyway, we see that C<903> ids correspond to creation of globs, twice
885 per glob - for glob name, and glob stringification magic.
887 Here are explanations for other I<Id>s above:
893 Creates bigger C<XPV*> structures. In the case above, it
894 creates 3 C<AV>s per subroutine, one for a list of lexical variable
895 names, one for a scratchpad (which contains lexical variables and
896 C<targets>), and one for the array of scratchpads needed for
899 It also creates a C<GV> and a C<CV> per subroutine, all called from
904 Creates a C array corresponding to the C<AV> of scratchpads and the
905 scratchpad itself. The first fake entry of this scratchpad is
906 created though the subroutine itself is not defined yet.
908 It also creates C arrays to keep data for the stash. This is one HV,
909 but it grows; thus, there are 4 big allocations: the big chunks are not
910 freed, but are kept as additional arenas for C<SV> allocations.
914 Creates a C<HEK> for the name of the glob for the subroutine. This
915 name is a key in a I<stash>.
917 Big allocations with this I<Id> correspond to allocations of new
918 arenas to keep C<HE>.
922 Creates a C<GP> for the glob for the subroutine.
926 Creates the C<MAGIC> for the glob for the subroutine.
930 Creates I<arenas> which keep SVs.
934 =head2 B<-DL> details
936 If Perl is run with B<-DL> option, then warn()s that start with `!'
937 behave specially. They print a list of I<categories> of memory
938 allocations, and statistics of allocations of different sizes for
941 If warn() string starts with
947 print changed categories only, print the differences in counts of allocations.
951 print grown categories only; print the absolute values of counts, and totals.
955 print nonempty categories, print the absolute values of counts and totals.
959 =head2 Limitations of B<-DL> statistics
961 If an extension or external library does not use the Perl API to
962 allocate memory, such allocations are not counted.