Commit | Line | Data |
055fd3a9 |
1 | =head1 NAME |
2 | |
3 | perldebguts - Guts of Perl debugging |
4 | |
5 | =head1 DESCRIPTION |
6 | |
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. |
11 | |
12 | =head1 Debugger Internals |
13 | |
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 |
4375e838 |
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 |
055fd3a9 |
18 | F<INSTALL> podpage in the Perl source tree. |
19 | |
20 | For example, whenever you call Perl's built-in C<caller> function |
21 | from the package DB, the arguments that the corresponding stack |
84880555 |
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 |
055fd3a9 |
24 | following additional features are enabled (cf. L<perlvar/$^P>): |
25 | |
13a2d996 |
26 | =over 4 |
055fd3a9 |
27 | |
28 | =item * |
29 | |
30 | Perl inserts the contents of C<$ENV{PERL5DB}> (or C<BEGIN {require |
31 | 'perl5db.pl'}> if not present) before the first line of your program. |
32 | |
33 | =item * |
34 | |
aa0b556f |
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 |
055fd3a9 |
37 | subroutines, or which are currently being executed. The $filename |
38 | for C<eval>ed strings looks like C<(eval 34)>. Code assertions |
8894c26d |
39 | in regexes look like C<(re_eval 19)>. |
40 | |
41 | Values in this array are magical in numeric context: they compare |
42 | equal to zero only if the line is not breakable. |
055fd3a9 |
43 | |
44 | =item * |
45 | |
aa0b556f |
46 | Each hash C<%{"_<$filename"}> contains breakpoints and actions keyed |
055fd3a9 |
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 |
8894c26d |
50 | C<"$break_condition\0$action">. |
055fd3a9 |
51 | |
52 | The same holds for evaluated strings that contain subroutines, or |
53 | which are currently being executed. The $filename for C<eval>ed strings |
54 | looks like C<(eval 34)> or C<(re_eval 19)>. |
55 | |
56 | =item * |
57 | |
aa0b556f |
58 | Each scalar C<${"_<$filename"}> contains C<"_<$filename">. This is |
055fd3a9 |
59 | also the case for evaluated strings that contain subroutines, or |
60 | which are currently being executed. The $filename for C<eval>ed |
61 | strings looks like C<(eval 34)> or C<(re_eval 19)>. |
62 | |
63 | =item * |
64 | |
65 | After each C<require>d file is compiled, but before it is executed, |
66 | C<DB::postponed(*{"_<$filename"})> is called if the subroutine |
67 | C<DB::postponed> exists. Here, the $filename is the expanded name of |
68 | the C<require>d file, as found in the values of %INC. |
69 | |
70 | =item * |
71 | |
72 | After each subroutine C<subname> is compiled, the existence of |
73 | C<$DB::postponed{subname}> is checked. If this key exists, |
74 | C<DB::postponed(subname)> is called if the C<DB::postponed> subroutine |
75 | also exists. |
76 | |
77 | =item * |
78 | |
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. |
83 | |
84 | =item * |
85 | |
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<< $^D & (1<<30) >> is true. |
92 | |
93 | =item * |
94 | |
95 | When execution of the program reaches a subroutine call, a call to |
96 | C<&DB::sub>(I<args>) is made instead, with C<$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.) |
99 | |
100 | =back |
101 | |
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. |
107 | |
108 | =head2 Writing Your Own Debugger |
109 | |
110 | The minimal working debugger consists of one line |
111 | |
112 | sub DB::DB {} |
113 | |
84880555 |
114 | which you could even fit into the C<PERL5DB> environment |
055fd3a9 |
115 | variable: |
116 | |
117 | $ PERL5DB="sub DB::DB {}" perl -d your-script |
118 | |
84880555 |
119 | although it doesn't do anything that tells you it's working... |
055fd3a9 |
120 | Another brief debugger, slightly more useful, could be created |
121 | with only the line: |
122 | |
123 | sub DB::DB {print ++$i; scalar <STDIN>} |
124 | |
125 | This debugger would print the sequential number of encountered |
126 | statement, and would wait for you to hit a newline before continuing. |
127 | |
128 | The following debugger is quite functional: |
129 | |
130 | { |
131 | package DB; |
132 | sub DB {} |
133 | sub sub {print ++$i, " $sub\n"; &$sub} |
134 | } |
135 | |
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 |
138 | package C<DB>. |
139 | |
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 |
143 | initialized. |
144 | |
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. |
148 | |
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. |
154 | |
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 |
106325ad |
160 | C<count> is missing). Each entry is reference to a hash with |
055fd3a9 |
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>. |
164 | |
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. |
168 | |
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. |
172 | |
173 | =head1 Frame Listing Output Examples |
174 | |
175 | The C<frame> option can be used to control the output of frame |
176 | information. For example, contrast this expression trace: |
177 | |
178 | $ perl -de 42 |
179 | Stack dump during die enabled outside of evals. |
180 | |
181 | Loading DB routines from perl5db.pl patch level 0.94 |
182 | Emacs support available. |
183 | |
184 | Enter h or `h h' for help. |
185 | |
186 | main::(-e:1): 0 |
187 | DB<1> sub foo { 14 } |
188 | |
189 | DB<2> sub bar { 3 } |
190 | |
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): |
195 | 42 |
196 | |
197 | with this one, once the C<O>ption C<frame=2> has been set: |
198 | |
199 | DB<4> O f=2 |
200 | frame = '2' |
201 | DB<5> t print foo() * bar() |
202 | 3: foo() * bar() |
203 | entering main::foo |
204 | 2: sub foo { 14 }; |
205 | exited main::foo |
206 | entering main::bar |
207 | 2: sub bar { 3 }; |
208 | exited main::bar |
209 | 42 |
210 | |
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. |
217 | |
218 | =over 4 |
219 | |
220 | =item 1 |
221 | |
222 | entering main::BEGIN |
223 | entering Config::BEGIN |
224 | Package lib/Exporter.pm. |
225 | Package lib/Carp.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 |
235 | |
236 | =item 2 |
237 | |
238 | entering main::BEGIN |
239 | entering Config::BEGIN |
240 | Package lib/Exporter.pm. |
241 | Package lib/Carp.pm. |
242 | exited Config::BEGIN |
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 |
250 | exited main::BEGIN |
251 | entering Config::myconfig |
252 | entering Config::FETCH |
253 | exited Config::FETCH |
254 | entering Config::FETCH |
255 | exited Config::FETCH |
256 | entering Config::FETCH |
257 | |
258 | =item 4 |
259 | |
260 | in $=main::BEGIN() from /dev/null:0 |
261 | in $=Config::BEGIN() from lib/Config.pm:2 |
262 | Package lib/Exporter.pm. |
263 | Package lib/Carp.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 |
275 | |
276 | =item 6 |
277 | |
278 | in $=main::BEGIN() from /dev/null:0 |
279 | in $=Config::BEGIN() from lib/Config.pm:2 |
280 | Package lib/Exporter.pm. |
281 | Package lib/Carp.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 |
299 | |
300 | =item 14 |
301 | |
302 | in $=main::BEGIN() from /dev/null:0 |
303 | in $=Config::BEGIN() from lib/Config.pm:2 |
304 | Package lib/Exporter.pm. |
305 | Package lib/Carp.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 |
320 | |
321 | =item 30 |
322 | |
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 $=Exporter::import('Config', 'myconfig', 'config_vars') from /dev/null:0 |
337 | scalar context return from Exporter::import: '' |
338 | |
339 | =back |
340 | |
341 | In all cases shown above, the line indentation shows the call tree. |
342 | If bit 2 of C<frame> is set, a line is printed on exit from a |
343 | subroutine as well. If bit 4 is set, the arguments are printed |
344 | along with the caller info. If bit 8 is set, the arguments are |
345 | printed even if they are tied or references. If bit 16 is set, the |
346 | return value is printed, too. |
347 | |
348 | When a package is compiled, a line like this |
349 | |
350 | Package lib/Carp.pm. |
351 | |
352 | is printed with proper indentation. |
353 | |
354 | =head1 Debugging regular expressions |
355 | |
356 | There are two ways to enable debugging output for regular expressions. |
357 | |
358 | If your perl is compiled with C<-DDEBUGGING>, you may use the |
359 | B<-Dr> flag on the command line. |
360 | |
361 | Otherwise, one can C<use re 'debug'>, which has effects at |
362 | compile time and run time. It is not lexically scoped. |
363 | |
364 | =head2 Compile-time output |
365 | |
366 | The debugging output at compile time looks like this: |
367 | |
1c102323 |
368 | Compiling REx `[bc]d(ef*g)+h[ij]k$' |
369 | size 45 Got 364 bytes for offset annotations. |
370 | first at 1 |
371 | rarest char g at 0 |
372 | rarest char d at 0 |
373 | 1: ANYOF[bc](12) |
374 | 12: EXACT <d>(14) |
375 | 14: CURLYX[0] {1,32767}(28) |
376 | 16: OPEN1(18) |
377 | 18: EXACT <e>(20) |
378 | 20: STAR(23) |
379 | 21: EXACT <f>(0) |
380 | 23: EXACT <g>(25) |
381 | 25: CLOSE1(27) |
382 | 27: WHILEM[1/1](0) |
383 | 28: NOTHING(29) |
384 | 29: EXACT <h>(31) |
385 | 31: ANYOF[ij](42) |
386 | 42: EXACT <k>(44) |
387 | 44: EOL(45) |
388 | 45: END(0) |
389 | anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating) |
390 | stclass `ANYOF[bc]' minlen 7 |
391 | Offsets: [45] |
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. |
055fd3a9 |
397 | |
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 |
1c102323 |
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. |
055fd3a9 |
403 | |
1c102323 |
404 | The |
405 | |
406 | anchored `de' at 1 floating `gh' at 3..2147483647 (checking floating) |
407 | stclass `ANYOF[bc]' minlen 7 |
408 | |
409 | line (split into two lines above) contains optimizer |
055fd3a9 |
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 |
1c102323 |
416 | C<first> I<id>) with a character class, and no string |
417 | shorter than 7 characters can possibly match. |
055fd3a9 |
418 | |
1c102323 |
419 | The fields of interest which may appear in this line are |
055fd3a9 |
420 | |
13a2d996 |
421 | =over 4 |
055fd3a9 |
422 | |
423 | =item C<anchored> I<STRING> C<at> I<POS> |
424 | |
425 | =item C<floating> I<STRING> C<at> I<POS1..POS2> |
426 | |
427 | See above. |
428 | |
429 | =item C<matching floating/anchored> |
430 | |
431 | Which substring to check first. |
432 | |
433 | =item C<minlen> |
434 | |
435 | The minimal length of the match. |
436 | |
437 | =item C<stclass> I<TYPE> |
438 | |
439 | Type of first matching node. |
440 | |
441 | =item C<noscan> |
442 | |
443 | Don't scan for the found substrings. |
444 | |
445 | =item C<isall> |
446 | |
1c102323 |
447 | Means that the optimizer information is all that the regular |
055fd3a9 |
448 | expression contains, and thus one does not need to enter the regex engine at |
449 | all. |
450 | |
451 | =item C<GPOS> |
452 | |
453 | Set if the pattern contains C<\G>. |
454 | |
455 | =item C<plus> |
456 | |
457 | Set if the pattern starts with a repeated char (as in C<x+y>). |
458 | |
459 | =item C<implicit> |
460 | |
461 | Set if the pattern starts with C<.*>. |
462 | |
463 | =item C<with eval> |
464 | |
465 | Set if the pattern contain eval-groups, such as C<(?{ code })> and |
466 | C<(??{ code })>. |
467 | |
468 | =item C<anchored(TYPE)> |
469 | |
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. |
472 | |
473 | =back |
474 | |
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'$>. |
477 | |
1c102323 |
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 |
055fd3a9 |
480 | is set, a call to the regex engine may be avoided even when the optimizer |
481 | found an appropriate place for the match. |
482 | |
1c102323 |
483 | Above the optimizer section is the list of I<nodes> of the compiled |
055fd3a9 |
484 | form of the regex. Each line has format |
485 | |
486 | C< >I<id>: I<TYPE> I<OPTIONAL-INFO> (I<next-id>) |
487 | |
488 | =head2 Types of nodes |
489 | |
490 | Here are the possible types, with short descriptions: |
491 | |
492 | # TYPE arg-description [num-args] [longjump-len] DESCRIPTION |
493 | |
494 | # Exit points |
495 | END no End of program. |
496 | SUCCEED no Return from a subroutine, basically. |
497 | |
498 | # Anchors: |
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. |
511 | |
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 |
526 | |
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. |
534 | # |
535 | BRANCH node Match this alternative, or the next... |
536 | |
537 | # BACK Normal "next" pointers all implicitly point forward; BACK |
538 | # exists to make loop structures possible. |
539 | # not used |
540 | BACK no Match "", "next" ptr points backward. |
541 | |
542 | # Literals |
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). |
546 | |
547 | # Do nothing |
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. |
551 | |
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. |
556 | # |
557 | STAR node Match this (simple) thing 0 or more times. |
558 | PLUS node Match this (simple) thing 1 or more times. |
559 | |
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. |
565 | |
566 | # This terminator creates a loop structure for CURLYX |
567 | WHILEM no Do curly processing and see if rest matches. |
568 | |
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. |
572 | |
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. |
576 | |
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. |
583 | |
584 | # Support for long regex |
585 | LONGJMP off 1 1 Jump far away. |
586 | BRANCHJ off 1 1 BRANCH with long offset. |
587 | |
588 | # The heavy worker |
589 | EVAL evl 1 Execute some Perl code. |
590 | |
591 | # Modifiers |
592 | MINMOD no Next operator is not greedy. |
593 | LOGICAL no Next opcode should set the flag only. |
594 | |
595 | # This is not used yet |
596 | RENUM off 1 1 Group with independently numbered parens. |
597 | |
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. |
601 | |
1c102323 |
602 | =for unprinted-credits |
603 | Next section M-J. Dominus (mjd-perl-patch+@plover.com) 20010421 |
604 | |
605 | Following the optimizer information is a dump of the offset/length |
606 | table, here split across several lines: |
607 | |
608 | Offsets: [45] |
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] |
613 | |
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]>. |
17c338f3 |
616 | Entries are numbered starting with 1, so entry #1 here is C<1[4]> and |
1c102323 |
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. |
629 | |
630 | C<0[0]> items indicate that there is no corresponding node. |
631 | |
055fd3a9 |
632 | =head2 Run-time output |
633 | |
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. |
637 | |
638 | If the regex engine was entered, the output may look like this: |
639 | |
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> |
663 | failed... |
664 | failed... |
665 | |
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 |
669 | |
670 | C< >I<STRING-OFFSET> <I<PRE-STRING>> <I<POST-STRING>> |I<ID>: I<TYPE> |
671 | |
672 | The I<TYPE> info is indented with respect to the backtracking level. |
673 | Other incidental information appears interspersed within. |
674 | |
675 | =head1 Debugging Perl memory usage |
676 | |
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 |
4375e838 |
681 | astonished. This is not absolutely true, but may provide a good |
055fd3a9 |
682 | grasp of what happens. |
683 | |
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 |
b9449ee0 |
690 | 20 bytes. A sloppy malloc(3) implementation can inflate these |
055fd3a9 |
691 | numbers dramatically. |
692 | |
693 | On the opposite end of the scale, a declaration like |
694 | |
695 | sub foo; |
696 | |
697 | may take up to 500 bytes of memory, depending on which release of Perl |
698 | you're running. |
699 | |
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 |
704 | on disk. |
705 | |
706 | There are two Perl-specific ways to analyze memory usage: |
707 | $ENV{PERL_DEBUG_MSTATS} and B<-DL> command-line switch. The first |
708 | is available only if Perl is compiled with Perl's malloc(); the |
709 | second only if Perl was built with C<-DDEBUGGING>. See the |
710 | instructions for how to do this in the F<INSTALL> podpage at |
711 | the top level of the Perl source tree. |
712 | |
713 | =head2 Using C<$ENV{PERL_DEBUG_MSTATS}> |
714 | |
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 |
4375e838 |
717 | usage statistics after compiling your code when C<< $ENV{PERL_DEBUG_MSTATS} |
055fd3a9 |
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: |
721 | |
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 |
725 | 437 61 36 0 5 |
726 | 60924 used: 125 137 161 55 7 8 6 16 2 0 1 |
727 | 74 109 304 84 20 |
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 |
731 | 315 162 39 42 11 |
732 | 175816 used: 265 176 1112 111 26 22 11 27 2 1 1 |
733 | 196 178 1066 798 39 |
734 | Total sbrk(): 215040/47:145. Odd ends: pad+heads+chain+tail: 0+2192+0+6144. |
735 | |
736 | It is possible to ask for such a statistic at arbitrary points in |
b9449ee0 |
737 | your execution using the mstat() function out of the standard |
055fd3a9 |
738 | Devel::Peek module. |
739 | |
740 | Here is some explanation of that format: |
741 | |
13a2d996 |
742 | =over 4 |
055fd3a9 |
743 | |
744 | =item C<buckets SMALLEST(APPROX)..GREATEST(APPROX)> |
745 | |
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. |
749 | |
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. |
755 | |
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. |
761 | |
762 | =item Free/Used |
763 | |
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". |
770 | |
4375e838 |
771 | For example, suppose under the previous example, the memory footprints |
055fd3a9 |
772 | were |
773 | |
774 | free: 8 16 32 64 128 256 512 1024 2048 4096 8192 |
775 | 4 12 24 48 80 |
776 | |
777 | With non-C<DEBUGGING> perl, the buckets starting from C<128> have |
d1be9408 |
778 | a 4-byte overhead, and thus an 8192-long bucket may take up to |
055fd3a9 |
779 | 8188-byte allocations. |
780 | |
781 | =item C<Total sbrk(): SBRKed/SBRKs:CONTINUOUS> |
782 | |
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. |
788 | |
789 | Memory allocated by external libraries is not counted. |
790 | |
791 | =item C<pad: 0> |
792 | |
793 | The amount of sbrk(2)ed memory needed to keep buckets aligned. |
794 | |
795 | =item C<heads: 2192> |
796 | |
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. |
800 | |
801 | =item C<chain: 0> |
802 | |
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. |
807 | |
808 | =item C<tail: 6144> |
809 | |
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 |
812 | never touched. |
813 | |
814 | =back |
815 | |
816 | =head2 Example of using B<-DL> switch |
817 | |
818 | Below we show how to analyse memory usage by |
819 | |
820 | do 'lib/auto/POSIX/autosplit.ix'; |
821 | |
822 | The file in question contains a header and 146 lines similar to |
823 | |
824 | sub getcwd; |
825 | |
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.) |
832 | |
833 | Here is the itemized list of Perl allocations performed during parsing |
834 | of this file: |
835 | |
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 . . . . . . . . |
847 | |
848 | |
849 | To see this list, insert two C<warn('!...')> statements around the call: |
850 | |
851 | warn('!'); |
852 | do 'lib/auto/POSIX/autosplit.ix'; |
853 | warn('!!! "after"'); |
854 | |
4375e838 |
855 | and run it with Perl's B<-DL> option. The first warn() will print |
055fd3a9 |
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. |
860 | |
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. |
872 | |
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 |
875 | set a C breakpoint |
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. |
883 | |
884 | Anyway, we see that C<903> ids correspond to creation of globs, twice |
885 | per glob - for glob name, and glob stringification magic. |
886 | |
887 | Here are explanations for other I<Id>s above: |
888 | |
13a2d996 |
889 | =over 4 |
055fd3a9 |
890 | |
891 | =item C<717> |
892 | |
4375e838 |
893 | Creates bigger C<XPV*> structures. In the case above, it |
055fd3a9 |
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 |
897 | recursion. |
898 | |
899 | It also creates a C<GV> and a C<CV> per subroutine, all called from |
900 | start_subparse(). |
901 | |
902 | =item C<002> |
903 | |
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. |
907 | |
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. |
911 | |
912 | =item C<054> |
913 | |
914 | Creates a C<HEK> for the name of the glob for the subroutine. This |
915 | name is a key in a I<stash>. |
916 | |
917 | Big allocations with this I<Id> correspond to allocations of new |
918 | arenas to keep C<HE>. |
919 | |
920 | =item C<602> |
921 | |
922 | Creates a C<GP> for the glob for the subroutine. |
923 | |
924 | =item C<702> |
925 | |
926 | Creates the C<MAGIC> for the glob for the subroutine. |
927 | |
928 | =item C<704> |
929 | |
930 | Creates I<arenas> which keep SVs. |
931 | |
932 | =back |
933 | |
934 | =head2 B<-DL> details |
935 | |
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 |
939 | these categories. |
940 | |
941 | If warn() string starts with |
942 | |
13a2d996 |
943 | =over 4 |
055fd3a9 |
944 | |
945 | =item C<!!!> |
946 | |
947 | print changed categories only, print the differences in counts of allocations. |
948 | |
949 | =item C<!!> |
950 | |
951 | print grown categories only; print the absolute values of counts, and totals. |
952 | |
953 | =item C<!> |
954 | |
955 | print nonempty categories, print the absolute values of counts and totals. |
956 | |
957 | =back |
958 | |
959 | =head2 Limitations of B<-DL> statistics |
960 | |
961 | If an extension or external library does not use the Perl API to |
962 | allocate memory, such allocations are not counted. |
963 | |
964 | =head1 SEE ALSO |
965 | |
966 | L<perldebug>, |
967 | L<perlguts>, |
968 | L<perlrun> |
969 | L<re>, |
970 | and |
fe854a6f |
971 | L<Devel::DProf>. |