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 @DB::args array. The
23 general mechanisms 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 is quite handy as contents of C<PERL5DB> environment
117 $ PERL5DB="sub DB::DB {}" perl -d your-script
119 Another brief debugger, slightly more useful, could be created
122 sub DB::DB {print ++$i; scalar <STDIN>}
124 This debugger would print the sequential number of encountered
125 statement, and would wait for you to hit a newline before continuing.
127 The following debugger is quite functional:
132 sub sub {print ++$i, " $sub\n"; &$sub}
135 It prints the sequential number of subroutine call and the name of the
136 called subroutine. Note that C<&DB::sub> should be compiled into the
139 At the start, the debugger reads your rc file (F<./.perldb> or
140 F<~/.perldb> under Unix), which can set important options. This file may
141 define a subroutine C<&afterinit> to be executed after the debugger is
144 After the rc file is read, the debugger reads the PERLDB_OPTS
145 environment variable and parses this as the remainder of a C<O ...>
146 line as one might enter at the debugger prompt.
148 The debugger also maintains magical internal variables, such as
149 C<@DB::dbline>, C<%DB::dbline>, which are aliases for
150 C<@{"::_<current_file"}> C<%{"::_<current_file"}>. Here C<current_file>
151 is the currently selected file, either explicitly chosen with the
152 debugger's C<f> command, or implicitly by flow of execution.
154 Some functions are provided to simplify customization. See
155 L<perldebug/"Options"> for description of options parsed by
156 C<DB::parse_options(string)>. The function C<DB::dump_trace(skip[,
157 count])> skips the specified number of frames and returns a list
158 containing information about the calling frames (all of them, if
159 C<count> is missing). Each entry is reference to a hash with
160 keys C<context> (either C<.>, C<$>, or C<@>), C<sub> (subroutine
161 name, or info about C<eval>), C<args> (C<undef> or a reference to
162 an array), C<file>, and C<line>.
164 The function C<DB::print_trace(FH, skip[, count[, short]])> prints
165 formatted info about caller frames. The last two functions may be
166 convenient as arguments to C<< < >>, C<< << >> commands.
168 Note that any variables and functions that are not documented in
169 this manpages (or in L<perldebug>) are considered for internal
170 use only, and as such are subject to change without notice.
172 =head1 Frame Listing Output Examples
174 The C<frame> option can be used to control the output of frame
175 information. For example, contrast this expression trace:
178 Stack dump during die enabled outside of evals.
180 Loading DB routines from perl5db.pl patch level 0.94
181 Emacs support available.
183 Enter h or `h h' for help.
190 DB<3> t print foo() * bar()
191 main::((eval 172):3): print foo() + bar();
192 main::foo((eval 168):2):
193 main::bar((eval 170):2):
196 with this one, once the C<O>ption C<frame=2> has been set:
200 DB<5> t print foo() * bar()
210 By way of demonstration, we present below a laborious listing
211 resulting from setting your C<PERLDB_OPTS> environment variable to
212 the value C<f=n N>, and running I<perl -d -V> from the command line.
213 Examples use various values of C<n> are shown to give you a feel
214 for the difference between settings. Long those it may be, this
215 is not a complete listing, but only excerpts.
222 entering Config::BEGIN
223 Package lib/Exporter.pm.
225 Package lib/Config.pm.
226 entering Config::TIEHASH
227 entering Exporter::import
228 entering Exporter::export
229 entering Config::myconfig
230 entering Config::FETCH
231 entering Config::FETCH
232 entering Config::FETCH
233 entering Config::FETCH
238 entering Config::BEGIN
239 Package lib/Exporter.pm.
242 Package lib/Config.pm.
243 entering Config::TIEHASH
244 exited Config::TIEHASH
245 entering Exporter::import
246 entering Exporter::export
247 exited Exporter::export
248 exited Exporter::import
250 entering Config::myconfig
251 entering Config::FETCH
253 entering Config::FETCH
255 entering Config::FETCH
259 in $=main::BEGIN() from /dev/null:0
260 in $=Config::BEGIN() from lib/Config.pm:2
261 Package lib/Exporter.pm.
263 Package lib/Config.pm.
264 in $=Config::TIEHASH('Config') from lib/Config.pm:644
265 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
266 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from li
267 in @=Config::myconfig() from /dev/null:0
268 in $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
269 in $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
270 in $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
271 in $=Config::FETCH(ref(Config), 'PERL_SUBVERSION') from lib/Config.pm:574
272 in $=Config::FETCH(ref(Config), 'osname') from lib/Config.pm:574
273 in $=Config::FETCH(ref(Config), 'osvers') from lib/Config.pm:574
277 in $=main::BEGIN() from /dev/null:0
278 in $=Config::BEGIN() from lib/Config.pm:2
279 Package lib/Exporter.pm.
281 out $=Config::BEGIN() from lib/Config.pm:0
282 Package lib/Config.pm.
283 in $=Config::TIEHASH('Config') from lib/Config.pm:644
284 out $=Config::TIEHASH('Config') from lib/Config.pm:644
285 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
286 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/
287 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/
288 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
289 out $=main::BEGIN() from /dev/null:0
290 in @=Config::myconfig() from /dev/null:0
291 in $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
292 out $=Config::FETCH(ref(Config), 'package') from lib/Config.pm:574
293 in $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
294 out $=Config::FETCH(ref(Config), 'baserev') from lib/Config.pm:574
295 in $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
296 out $=Config::FETCH(ref(Config), 'PERL_VERSION') from lib/Config.pm:574
297 in $=Config::FETCH(ref(Config), 'PERL_SUBVERSION') from lib/Config.pm:574
301 in $=main::BEGIN() from /dev/null:0
302 in $=Config::BEGIN() from lib/Config.pm:2
303 Package lib/Exporter.pm.
305 out $=Config::BEGIN() from lib/Config.pm:0
306 Package lib/Config.pm.
307 in $=Config::TIEHASH('Config') from lib/Config.pm:644
308 out $=Config::TIEHASH('Config') from lib/Config.pm:644
309 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
310 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/E
311 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/E
312 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
313 out $=main::BEGIN() from /dev/null:0
314 in @=Config::myconfig() from /dev/null:0
315 in $=Config::FETCH('Config=HASH(0x1aa444)', 'package') from lib/Config.pm:574
316 out $=Config::FETCH('Config=HASH(0x1aa444)', 'package') from lib/Config.pm:574
317 in $=Config::FETCH('Config=HASH(0x1aa444)', 'baserev') from lib/Config.pm:574
318 out $=Config::FETCH('Config=HASH(0x1aa444)', 'baserev') from lib/Config.pm:574
322 in $=CODE(0x15eca4)() from /dev/null:0
323 in $=CODE(0x182528)() from lib/Config.pm:2
324 Package lib/Exporter.pm.
325 out $=CODE(0x182528)() from lib/Config.pm:0
326 scalar context return from CODE(0x182528): undef
327 Package lib/Config.pm.
328 in $=Config::TIEHASH('Config') from lib/Config.pm:628
329 out $=Config::TIEHASH('Config') from lib/Config.pm:628
330 scalar context return from Config::TIEHASH: empty hash
331 in $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
332 in $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/Exporter.pm:171
333 out $=Exporter::export('Config', 'main', 'myconfig', 'config_vars') from lib/Exporter.pm:171
334 scalar context return from Exporter::export: ''
335 out $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0
336 scalar context return from Exporter::import: ''
340 In all cases shown above, the line indentation shows the call tree.
341 If bit 2 of C<frame> is set, a line is printed on exit from a
342 subroutine as well. If bit 4 is set, the arguments are printed
343 along with the caller info. If bit 8 is set, the arguments are
344 printed even if they are tied or references. If bit 16 is set, the
345 return value is printed, too.
347 When a package is compiled, a line like this
351 is printed with proper indentation.
353 =head1 Debugging regular expressions
355 There are two ways to enable debugging output for regular expressions.
357 If your perl is compiled with C<-DDEBUGGING>, you may use the
358 B<-Dr> flag on the command line.
360 Otherwise, one can C<use re 'debug'>, which has effects at
361 compile time and run time. It is not lexically scoped.
363 =head2 Compile-time output
365 The debugging output at compile time looks like this:
367 Compiling REx `[bc]d(ef*g)+h[ij]k$'
368 size 45 Got 364 bytes for offset annotations.
374 14: CURLYX[0] {1,32767}(28)
388 anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating)
389 stclass `ANYOF[bc]' minlen 7
391 1[4] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 5[1]
392 0[0] 12[1] 0[0] 6[1] 0[0] 7[1] 0[0] 9[1] 8[1] 0[0] 10[1] 0[0]
393 11[1] 0[0] 12[0] 12[0] 13[1] 0[0] 14[4] 0[0] 0[0] 0[0] 0[0]
394 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 18[1] 0[0] 19[1] 20[0]
395 Omitting $` $& $' support.
397 The first line shows the pre-compiled form of the regex. The second
398 shows the size of the compiled form (in arbitrary units, usually
399 4-byte words) and the total number of bytes allocated for the
400 offset/length table, usually 4+C<size>*8. The next line shows the
401 label I<id> of the first node that does a match.
405 anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating)
406 stclass `ANYOF[bc]' minlen 7
408 line (split into two lines above) contains optimizer
409 information. In the example shown, the optimizer found that the match
410 should contain a substring C<de> at offset 1, plus substring C<gh>
411 at some offset between 3 and infinity. Moreover, when checking for
412 these substrings (to abandon impossible matches quickly), Perl will check
413 for the substring C<gh> before checking for the substring C<de>. The
414 optimizer may also use the knowledge that the match starts (at the
415 C<first> I<id>) with a character class, and no string
416 shorter than 7 characters can possibly match.
418 The fields of interest which may appear in this line are
422 =item C<anchored> I<STRING> C<at> I<POS>
424 =item C<floating> I<STRING> C<at> I<POS1..POS2>
428 =item C<matching floating/anchored>
430 Which substring to check first.
434 The minimal length of the match.
436 =item C<stclass> I<TYPE>
438 Type of first matching node.
442 Don't scan for the found substrings.
446 Means that the optimizer information is all that the regular
447 expression contains, and thus one does not need to enter the regex engine at
452 Set if the pattern contains C<\G>.
456 Set if the pattern starts with a repeated char (as in C<x+y>).
460 Set if the pattern starts with C<.*>.
464 Set if the pattern contain eval-groups, such as C<(?{ code })> and
467 =item C<anchored(TYPE)>
469 If the pattern may match only at a handful of places, (with C<TYPE>
470 being C<BOL>, C<MBOL>, or C<GPOS>. See the table below.
474 If a substring is known to match at end-of-line only, it may be
475 followed by C<$>, as in C<floating `k'$>.
477 The optimizer-specific information is used to avoid entering (a slow) regex
478 engine on strings that will not definitely match. If the C<isall> flag
479 is set, a call to the regex engine may be avoided even when the optimizer
480 found an appropriate place for the match.
482 Above the optimizer section is the list of I<nodes> of the compiled
483 form of the regex. Each line has format
485 C< >I<id>: I<TYPE> I<OPTIONAL-INFO> (I<next-id>)
487 =head2 Types of nodes
489 Here are the possible types, with short descriptions:
491 # TYPE arg-description [num-args] [longjump-len] DESCRIPTION
494 END no End of program.
495 SUCCEED no Return from a subroutine, basically.
498 BOL no Match "" at beginning of line.
499 MBOL no Same, assuming multiline.
500 SBOL no Same, assuming singleline.
501 EOS no Match "" at end of string.
502 EOL no Match "" at end of line.
503 MEOL no Same, assuming multiline.
504 SEOL no Same, assuming singleline.
505 BOUND no Match "" at any word boundary
506 BOUNDL no Match "" at any word boundary
507 NBOUND no Match "" at any word non-boundary
508 NBOUNDL no Match "" at any word non-boundary
509 GPOS no Matches where last m//g left off.
511 # [Special] alternatives
512 ANY no Match any one character (except newline).
513 SANY no Match any one character.
514 ANYOF sv Match character in (or not in) this class.
515 ALNUM no Match any alphanumeric character
516 ALNUML no Match any alphanumeric char in locale
517 NALNUM no Match any non-alphanumeric character
518 NALNUML no Match any non-alphanumeric char in locale
519 SPACE no Match any whitespace character
520 SPACEL no Match any whitespace char in locale
521 NSPACE no Match any non-whitespace character
522 NSPACEL no Match any non-whitespace char in locale
523 DIGIT no Match any numeric character
524 NDIGIT no Match any non-numeric character
526 # BRANCH The set of branches constituting a single choice are hooked
527 # together with their "next" pointers, since precedence prevents
528 # anything being concatenated to any individual branch. The
529 # "next" pointer of the last BRANCH in a choice points to the
530 # thing following the whole choice. This is also where the
531 # final "next" pointer of each individual branch points; each
532 # branch starts with the operand node of a BRANCH node.
534 BRANCH node Match this alternative, or the next...
536 # BACK Normal "next" pointers all implicitly point forward; BACK
537 # exists to make loop structures possible.
539 BACK no Match "", "next" ptr points backward.
542 EXACT sv Match this string (preceded by length).
543 EXACTF sv Match this string, folded (prec. by length).
544 EXACTFL sv Match this string, folded in locale (w/len).
547 NOTHING no Match empty string.
548 # A variant of above which delimits a group, thus stops optimizations
549 TAIL no Match empty string. Can jump here from outside.
551 # STAR,PLUS '?', and complex '*' and '+', are implemented as circular
552 # BRANCH structures using BACK. Simple cases (one character
553 # per match) are implemented with STAR and PLUS for speed
554 # and to minimize recursive plunges.
556 STAR node Match this (simple) thing 0 or more times.
557 PLUS node Match this (simple) thing 1 or more times.
559 CURLY sv 2 Match this simple thing {n,m} times.
560 CURLYN no 2 Match next-after-this simple thing
561 # {n,m} times, set parens.
562 CURLYM no 2 Match this medium-complex thing {n,m} times.
563 CURLYX sv 2 Match this complex thing {n,m} times.
565 # This terminator creates a loop structure for CURLYX
566 WHILEM no Do curly processing and see if rest matches.
568 # OPEN,CLOSE,GROUPP ...are numbered at compile time.
569 OPEN num 1 Mark this point in input as start of #n.
570 CLOSE num 1 Analogous to OPEN.
572 REF num 1 Match some already matched string
573 REFF num 1 Match already matched string, folded
574 REFFL num 1 Match already matched string, folded in loc.
576 # grouping assertions
577 IFMATCH off 1 2 Succeeds if the following matches.
578 UNLESSM off 1 2 Fails if the following matches.
579 SUSPEND off 1 1 "Independent" sub-regex.
580 IFTHEN off 1 1 Switch, should be preceded by switcher .
581 GROUPP num 1 Whether the group matched.
583 # Support for long regex
584 LONGJMP off 1 1 Jump far away.
585 BRANCHJ off 1 1 BRANCH with long offset.
588 EVAL evl 1 Execute some Perl code.
591 MINMOD no Next operator is not greedy.
592 LOGICAL no Next opcode should set the flag only.
594 # This is not used yet
595 RENUM off 1 1 Group with independently numbered parens.
597 # This is not really a node, but an optimized away piece of a "long" node.
598 # To simplify debugging output, we mark it as if it were a node
599 OPTIMIZED off Placeholder for dump.
601 =for unprinted-credits
602 Next section M-J. Dominus (mjd-perl-patch+@plover.com) 20010421
604 Following the optimizer information is a dump of the offset/length
605 table, here split across several lines:
608 1[4] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 5[1]
609 0[0] 12[1] 0[0] 6[1] 0[0] 7[1] 0[0] 9[1] 8[1] 0[0] 10[1] 0[0]
610 11[1] 0[0] 12[0] 12[0] 13[1] 0[0] 14[4] 0[0] 0[0] 0[0] 0[0]
611 0[0] 0[0] 0[0] 0[0] 0[0] 0[0] 18[1] 0[0] 19[1] 20[0]
613 The first line here indicates that the offset/length table contains 45
614 entries. Each entry is a pair of integers, denoted by C<offset[length]>.
615 Entries are numbered starting with, so entry #1 here is C<1[4]> and
616 entry #12 is C<5[1]>. C<1[4]> indicates that the node labeled C<1:>
617 (the C<1: ANYOF[bc]>) begins at character position 1 in the
618 pre-compiled form of the regex, and has a length of 4 characters.
619 C<5[1]> in position 12
620 indicates that the node labeled C<12:>
621 (the C<< 12: EXACT <d> >>) begins at character position 5 in the
622 pre-compiled form of the regex, and has a length of 1 character.
623 C<12[1]> in position 14
624 indicates that the node labeled C<14:>
625 (the C<< 14: CURLYX[0] {1,32767} >>) begins at character position 12 in the
626 pre-compiled form of the regex, and has a length of 1 character---that
627 is, it corresponds to the C<+> symbol in the precompiled regex.
629 C<0[0]> items indicate that there is no corresponding node.
631 =head2 Run-time output
633 First of all, when doing a match, one may get no run-time output even
634 if debugging is enabled. This means that the regex engine was never
635 entered and that all of the job was therefore done by the optimizer.
637 If the regex engine was entered, the output may look like this:
639 Matching `[bc]d(ef*g)+h[ij]k$' against `abcdefg__gh__'
640 Setting an EVAL scope, savestack=3
641 2 <ab> <cdefg__gh_> | 1: ANYOF
642 3 <abc> <defg__gh_> | 11: EXACT <d>
643 4 <abcd> <efg__gh_> | 13: CURLYX {1,32767}
644 4 <abcd> <efg__gh_> | 26: WHILEM
645 0 out of 1..32767 cc=effff31c
646 4 <abcd> <efg__gh_> | 15: OPEN1
647 4 <abcd> <efg__gh_> | 17: EXACT <e>
648 5 <abcde> <fg__gh_> | 19: STAR
649 EXACT <f> can match 1 times out of 32767...
650 Setting an EVAL scope, savestack=3
651 6 <bcdef> <g__gh__> | 22: EXACT <g>
652 7 <bcdefg> <__gh__> | 24: CLOSE1
653 7 <bcdefg> <__gh__> | 26: WHILEM
654 1 out of 1..32767 cc=effff31c
655 Setting an EVAL scope, savestack=12
656 7 <bcdefg> <__gh__> | 15: OPEN1
657 7 <bcdefg> <__gh__> | 17: EXACT <e>
658 restoring \1 to 4(4)..7
659 failed, try continuation...
660 7 <bcdefg> <__gh__> | 27: NOTHING
661 7 <bcdefg> <__gh__> | 28: EXACT <h>
665 The most significant information in the output is about the particular I<node>
666 of the compiled regex that is currently being tested against the target string.
667 The format of these lines is
669 C< >I<STRING-OFFSET> <I<PRE-STRING>> <I<POST-STRING>> |I<ID>: I<TYPE>
671 The I<TYPE> info is indented with respect to the backtracking level.
672 Other incidental information appears interspersed within.
674 =head1 Debugging Perl memory usage
676 Perl is a profligate wastrel when it comes to memory use. There
677 is a saying that to estimate memory usage of Perl, assume a reasonable
678 algorithm for memory allocation, multiply that estimate by 10, and
679 while you still may miss the mark, at least you won't be quite so
680 astonished. This is not absolutely true, but may provide a good
681 grasp of what happens.
683 Assume that an integer cannot take less than 20 bytes of memory, a
684 float cannot take less than 24 bytes, a string cannot take less
685 than 32 bytes (all these examples assume 32-bit architectures, the
686 result are quite a bit worse on 64-bit architectures). If a variable
687 is accessed in two of three different ways (which require an integer,
688 a float, or a string), the memory footprint may increase yet another
689 20 bytes. A sloppy malloc(3) implementation can inflate these
690 numbers dramatically.
692 On the opposite end of the scale, a declaration like
696 may take up to 500 bytes of memory, depending on which release of Perl
699 Anecdotal estimates of source-to-compiled code bloat suggest an
700 eightfold increase. This means that the compiled form of reasonable
701 (normally commented, properly indented etc.) code will take
702 about eight times more space in memory than the code took
705 There are two Perl-specific ways to analyze memory usage:
706 $ENV{PERL_DEBUG_MSTATS} and B<-DL> command-line switch. The first
707 is available only if Perl is compiled with Perl's malloc(); the
708 second only if Perl was built with C<-DDEBUGGING>. See the
709 instructions for how to do this in the F<INSTALL> podpage at
710 the top level of the Perl source tree.
712 =head2 Using C<$ENV{PERL_DEBUG_MSTATS}>
714 If your perl is using Perl's malloc() and was compiled with the
715 necessary switches (this is the default), then it will print memory
716 usage statistics after compiling your code when C<< $ENV{PERL_DEBUG_MSTATS}
717 > 1 >>, and before termination of the program when C<<
718 $ENV{PERL_DEBUG_MSTATS} >= 1 >>. The report format is similar to
719 the following example:
721 $ PERL_DEBUG_MSTATS=2 perl -e "require Carp"
722 Memory allocation statistics after compilation: (buckets 4(4)..8188(8192)
723 14216 free: 130 117 28 7 9 0 2 2 1 0 0
725 60924 used: 125 137 161 55 7 8 6 16 2 0 1
727 Total sbrk(): 77824/21:119. Odd ends: pad+heads+chain+tail: 0+636+0+2048.
728 Memory allocation statistics after execution: (buckets 4(4)..8188(8192)
729 30888 free: 245 78 85 13 6 2 1 3 2 0 1
731 175816 used: 265 176 1112 111 26 22 11 27 2 1 1
733 Total sbrk(): 215040/47:145. Odd ends: pad+heads+chain+tail: 0+2192+0+6144.
735 It is possible to ask for such a statistic at arbitrary points in
736 your execution using the mstat() function out of the standard
739 Here is some explanation of that format:
743 =item C<buckets SMALLEST(APPROX)..GREATEST(APPROX)>
745 Perl's malloc() uses bucketed allocations. Every request is rounded
746 up to the closest bucket size available, and a bucket is taken from
747 the pool of buckets of that size.
749 The line above describes the limits of buckets currently in use.
750 Each bucket has two sizes: memory footprint and the maximal size
751 of user data that can fit into this bucket. Suppose in the above
752 example that the smallest bucket were size 4. The biggest bucket
753 would have usable size 8188, and the memory footprint would be 8192.
755 In a Perl built for debugging, some buckets may have negative usable
756 size. This means that these buckets cannot (and will not) be used.
757 For larger buckets, the memory footprint may be one page greater
758 than a power of 2. If so, case the corresponding power of two is
759 printed in the C<APPROX> field above.
763 The 1 or 2 rows of numbers following that correspond to the number
764 of buckets of each size between C<SMALLEST> and C<GREATEST>. In
765 the first row, the sizes (memory footprints) of buckets are powers
766 of two--or possibly one page greater. In the second row, if present,
767 the memory footprints of the buckets are between the memory footprints
768 of two buckets "above".
770 For example, suppose under the previous example, the memory footprints
773 free: 8 16 32 64 128 256 512 1024 2048 4096 8192
776 With non-C<DEBUGGING> perl, the buckets starting from C<128> have
777 a 4-byte overhead, and thus a 8192-long bucket may take up to
778 8188-byte allocations.
780 =item C<Total sbrk(): SBRKed/SBRKs:CONTINUOUS>
782 The first two fields give the total amount of memory perl sbrk(2)ed
783 (ess-broken? :-) and number of sbrk(2)s used. The third number is
784 what perl thinks about continuity of returned chunks. So long as
785 this number is positive, malloc() will assume that it is probable
786 that sbrk(2) will provide continuous memory.
788 Memory allocated by external libraries is not counted.
792 The amount of sbrk(2)ed memory needed to keep buckets aligned.
796 Although memory overhead of bigger buckets is kept inside the bucket, for
797 smaller buckets, it is kept in separate areas. This field gives the
798 total size of these areas.
802 malloc() may want to subdivide a bigger bucket into smaller buckets.
803 If only a part of the deceased bucket is left unsubdivided, the rest
804 is kept as an element of a linked list. This field gives the total
805 size of these chunks.
809 To minimize the number of sbrk(2)s, malloc() asks for more memory. This
810 field gives the size of the yet unused part, which is sbrk(2)ed, but
815 =head2 Example of using B<-DL> switch
817 Below we show how to analyse memory usage by
819 do 'lib/auto/POSIX/autosplit.ix';
821 The file in question contains a header and 146 lines similar to
825 B<WARNING>: The discussion below supposes 32-bit architecture. In
826 newer releases of Perl, memory usage of the constructs discussed
827 here is greatly improved, but the story discussed below is a real-life
828 story. This story is mercilessly terse, and assumes rather more than cursory
829 knowledge of Perl internals. Type space to continue, `q' to quit.
830 (Actually, you just want to skip to the next section.)
832 Here is the itemized list of Perl allocations performed during parsing
835 !!! "after" at test.pl line 3.
836 Id subtot 4 8 12 16 20 24 28 32 36 40 48 56 64 72 80 80+
837 0 02 13752 . . . . 294 . . . . . . . . . . 4
838 0 54 5545 . . 8 124 16 . . . 1 1 . . . . . 3
839 5 05 32 . . . . . . . 1 . . . . . . . .
840 6 02 7152 . . . . . . . . . . 149 . . . . .
841 7 02 3600 . . . . . 150 . . . . . . . . . .
842 7 03 64 . -1 . 1 . . 2 . . . . . . . . .
843 7 04 7056 . . . . . . . . . . . . . . . 7
844 7 17 38404 . . . . . . . 1 . . 442 149 . . 147 .
845 9 03 2078 17 249 32 . . . . 2 . . . . . . . .
848 To see this list, insert two C<warn('!...')> statements around the call:
851 do 'lib/auto/POSIX/autosplit.ix';
854 and run it with Perl's B<-DL> option. The first warn() will print
855 memory allocation info before parsing the file and will memorize
856 the statistics at this point (we ignore what it prints). The second
857 warn() prints increments with respect to these memorized data. This
858 is the printout shown above.
860 Different I<Id>s on the left correspond to different subsystems of
861 the perl interpreter. They are just the first argument given to
862 the perl memory allocation API named New(). To find what C<9 03>
863 means, just B<grep> the perl source for C<903>. You'll find it in
864 F<util.c>, function savepvn(). (I know, you wonder why we told you
865 to B<grep> and then gave away the answer. That's because grepping
866 the source is good for the soul.) This function is used to store
867 a copy of an existing chunk of memory. Using a C debugger, one can
868 see that the function was called either directly from gv_init() or
869 via sv_magic(), and that gv_init() is called from gv_fetchpv()--which
870 was itself called from newSUB(). Please stop to catch your breath now.
872 B<NOTE>: To reach this point in the debugger and skip the calls to
873 savepvn() during the compilation of the main program, you should
875 in Perl_warn(), continue until this point is reached, and I<then> set
876 a C breakpoint in Perl_savepvn(). Note that you may need to skip a
877 handful of Perl_savepvn() calls that do not correspond to mass production
878 of CVs (there are more C<903> allocations than 146 similar lines of
879 F<lib/auto/POSIX/autosplit.ix>). Note also that C<Perl_> prefixes are
880 added by macroization code in perl header files to avoid conflicts
881 with external libraries.
883 Anyway, we see that C<903> ids correspond to creation of globs, twice
884 per glob - for glob name, and glob stringification magic.
886 Here are explanations for other I<Id>s above:
892 Creates bigger C<XPV*> structures. In the case above, it
893 creates 3 C<AV>s per subroutine, one for a list of lexical variable
894 names, one for a scratchpad (which contains lexical variables and
895 C<targets>), and one for the array of scratchpads needed for
898 It also creates a C<GV> and a C<CV> per subroutine, all called from
903 Creates a C array corresponding to the C<AV> of scratchpads and the
904 scratchpad itself. The first fake entry of this scratchpad is
905 created though the subroutine itself is not defined yet.
907 It also creates C arrays to keep data for the stash. This is one HV,
908 but it grows; thus, there are 4 big allocations: the big chunks are not
909 freed, but are kept as additional arenas for C<SV> allocations.
913 Creates a C<HEK> for the name of the glob for the subroutine. This
914 name is a key in a I<stash>.
916 Big allocations with this I<Id> correspond to allocations of new
917 arenas to keep C<HE>.
921 Creates a C<GP> for the glob for the subroutine.
925 Creates the C<MAGIC> for the glob for the subroutine.
929 Creates I<arenas> which keep SVs.
933 =head2 B<-DL> details
935 If Perl is run with B<-DL> option, then warn()s that start with `!'
936 behave specially. They print a list of I<categories> of memory
937 allocations, and statistics of allocations of different sizes for
940 If warn() string starts with
946 print changed categories only, print the differences in counts of allocations.
950 print grown categories only; print the absolute values of counts, and totals.
954 print nonempty categories, print the absolute values of counts and totals.
958 =head2 Limitations of B<-DL> statistics
960 If an extension or external library does not use the Perl API to
961 allocate memory, such allocations are not counted.