3 perlhack - How to hack at the Perl internals
7 This document attempts to explain how Perl development takes place,
8 and ends with some suggestions for people wanting to become bona fide
11 The perl5-porters mailing list is where the Perl standard distribution
12 is maintained and developed. The list can get anywhere from 10 to 150
13 messages a day, depending on the heatedness of the debate. Most days
14 there are two or three patches, extensions, features, or bugs being
17 A searchable archive of the list is at either:
19 http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/
23 http://archive.develooper.com/perl5-porters@perl.org/
25 List subscribers (the porters themselves) come in several flavours.
26 Some are quiet curious lurkers, who rarely pitch in and instead watch
27 the ongoing development to ensure they're forewarned of new changes or
28 features in Perl. Some are representatives of vendors, who are there
29 to make sure that Perl continues to compile and work on their
30 platforms. Some patch any reported bug that they know how to fix,
31 some are actively patching their pet area (threads, Win32, the regexp
32 engine), while others seem to do nothing but complain. In other
33 words, it's your usual mix of technical people.
35 Over this group of porters presides Larry Wall. He has the final word
36 in what does and does not change in the Perl language. Various
37 releases of Perl are shepherded by a "pumpking", a porter
38 responsible for gathering patches, deciding on a patch-by-patch,
39 feature-by-feature basis what will and will not go into the release.
40 For instance, Gurusamy Sarathy was the pumpking for the 5.6 release of
41 Perl, and Jarkko Hietaniemi was the pumpking for the 5.8 release, and
42 Rafael Garcia-Suarez holds the pumpking crown for the 5.10 release.
44 In addition, various people are pumpkings for different things. For
45 instance, Andy Dougherty and Jarkko Hietaniemi did a grand job as the
46 I<Configure> pumpkin up till the 5.8 release. For the 5.10 release
47 H.Merijn Brand took over.
49 Larry sees Perl development along the lines of the US government:
50 there's the Legislature (the porters), the Executive branch (the
51 pumpkings), and the Supreme Court (Larry). The legislature can
52 discuss and submit patches to the executive branch all they like, but
53 the executive branch is free to veto them. Rarely, the Supreme Court
54 will side with the executive branch over the legislature, or the
55 legislature over the executive branch. Mostly, however, the
56 legislature and the executive branch are supposed to get along and
57 work out their differences without impeachment or court cases.
59 You might sometimes see reference to Rule 1 and Rule 2. Larry's power
60 as Supreme Court is expressed in The Rules:
66 Larry is always by definition right about how Perl should behave.
67 This means he has final veto power on the core functionality.
71 Larry is allowed to change his mind about any matter at a later date,
72 regardless of whether he previously invoked Rule 1.
76 Got that? Larry is always right, even when he was wrong. It's rare
77 to see either Rule exercised, but they are often alluded to.
79 New features and extensions to the language are contentious, because
80 the criteria used by the pumpkings, Larry, and other porters to decide
81 which features should be implemented and incorporated are not codified
82 in a few small design goals as with some other languages. Instead,
83 the heuristics are flexible and often difficult to fathom. Here is
84 one person's list, roughly in decreasing order of importance, of
85 heuristics that new features have to be weighed against:
89 =item Does concept match the general goals of Perl?
91 These haven't been written anywhere in stone, but one approximation
94 1. Keep it fast, simple, and useful.
95 2. Keep features/concepts as orthogonal as possible.
96 3. No arbitrary limits (platforms, data sizes, cultures).
97 4. Keep it open and exciting to use/patch/advocate Perl everywhere.
98 5. Either assimilate new technologies, or build bridges to them.
100 =item Where is the implementation?
102 All the talk in the world is useless without an implementation. In
103 almost every case, the person or people who argue for a new feature
104 will be expected to be the ones who implement it. Porters capable
105 of coding new features have their own agendas, and are not available
106 to implement your (possibly good) idea.
108 =item Backwards compatibility
110 It's a cardinal sin to break existing Perl programs. New warnings are
111 contentious--some say that a program that emits warnings is not
112 broken, while others say it is. Adding keywords has the potential to
113 break programs, changing the meaning of existing token sequences or
114 functions might break programs.
116 =item Could it be a module instead?
118 Perl 5 has extension mechanisms, modules and XS, specifically to avoid
119 the need to keep changing the Perl interpreter. You can write modules
120 that export functions, you can give those functions prototypes so they
121 can be called like built-in functions, you can even write XS code to
122 mess with the runtime data structures of the Perl interpreter if you
123 want to implement really complicated things. If it can be done in a
124 module instead of in the core, it's highly unlikely to be added.
126 =item Is the feature generic enough?
128 Is this something that only the submitter wants added to the language,
129 or would it be broadly useful? Sometimes, instead of adding a feature
130 with a tight focus, the porters might decide to wait until someone
131 implements the more generalized feature. For instance, instead of
132 implementing a "delayed evaluation" feature, the porters are waiting
133 for a macro system that would permit delayed evaluation and much more.
135 =item Does it potentially introduce new bugs?
137 Radical rewrites of large chunks of the Perl interpreter have the
138 potential to introduce new bugs. The smaller and more localized the
141 =item Does it preclude other desirable features?
143 A patch is likely to be rejected if it closes off future avenues of
144 development. For instance, a patch that placed a true and final
145 interpretation on prototypes is likely to be rejected because there
146 are still options for the future of prototypes that haven't been
149 =item Is the implementation robust?
151 Good patches (tight code, complete, correct) stand more chance of
152 going in. Sloppy or incorrect patches might be placed on the back
153 burner until the pumpking has time to fix, or might be discarded
154 altogether without further notice.
156 =item Is the implementation generic enough to be portable?
158 The worst patches make use of a system-specific features. It's highly
159 unlikely that non-portable additions to the Perl language will be
162 =item Is the implementation tested?
164 Patches which change behaviour (fixing bugs or introducing new features)
165 must include regression tests to verify that everything works as expected.
166 Without tests provided by the original author, how can anyone else changing
167 perl in the future be sure that they haven't unwittingly broken the behaviour
168 the patch implements? And without tests, how can the patch's author be
169 confident that his/her hard work put into the patch won't be accidentally
170 thrown away by someone in the future?
172 =item Is there enough documentation?
174 Patches without documentation are probably ill-thought out or
175 incomplete. Nothing can be added without documentation, so submitting
176 a patch for the appropriate manpages as well as the source code is
179 =item Is there another way to do it?
181 Larry said "Although the Perl Slogan is I<There's More Than One Way
182 to Do It>, I hesitate to make 10 ways to do something". This is a
183 tricky heuristic to navigate, though--one man's essential addition is
184 another man's pointless cruft.
186 =item Does it create too much work?
188 Work for the pumpking, work for Perl programmers, work for module
189 authors, ... Perl is supposed to be easy.
191 =item Patches speak louder than words
193 Working code is always preferred to pie-in-the-sky ideas. A patch to
194 add a feature stands a much higher chance of making it to the language
195 than does a random feature request, no matter how fervently argued the
196 request might be. This ties into "Will it be useful?", as the fact
197 that someone took the time to make the patch demonstrates a strong
198 desire for the feature.
202 If you're on the list, you might hear the word "core" bandied
203 around. It refers to the standard distribution. "Hacking on the
204 core" means you're changing the C source code to the Perl
205 interpreter. "A core module" is one that ships with Perl.
207 =head2 Keeping in sync
209 The source code to the Perl interpreter, in its different versions, is
210 kept in a repository managed by the git revision control system. The
211 pumpkings and a few others have write access to the repository to check in
214 How to clone and use the git perl repository is described in L<perlrepository>.
216 You can also choose to use rsync to get a copy of the current source tree
217 for the bleadperl branch and all maintenance branches :
219 $ rsync -avz rsync://perl5.git.perl.org/APC/perl-current .
220 $ rsync -avz rsync://perl5.git.perl.org/APC/perl-5.10.x .
221 $ rsync -avz rsync://perl5.git.perl.org/APC/perl-5.8.x .
222 $ rsync -avz rsync://perl5.git.perl.org/APC/perl-5.6.x .
223 $ rsync -avz rsync://perl5.git.perl.org/APC/perl-5.005xx .
225 (Add the C<--delete> option to remove leftover files)
227 You may also want to subscribe to the perl5-changes mailing list to
228 receive a copy of each patch that gets submitted to the maintenance
229 and development "branches" of the perl repository. See
230 http://lists.perl.org/ for subscription information.
232 If you are a member of the perl5-porters mailing list, it is a good
233 thing to keep in touch with the most recent changes. If not only to
234 verify if what you would have posted as a bug report isn't already
235 solved in the most recent available perl development branch, also
236 known as perl-current, bleading edge perl, bleedperl or bleadperl.
238 Needless to say, the source code in perl-current is usually in a perpetual
239 state of evolution. You should expect it to be very buggy. Do B<not> use
240 it for any purpose other than testing and development.
242 =head2 Perlbug administration
244 There is a single remote administrative interface for modifying bug status,
245 category, open issues etc. using the B<RT> bugtracker system, maintained
246 by Robert Spier. Become an administrator, and close any bugs you can get
247 your sticky mitts on:
249 http://bugs.perl.org/
251 To email the bug system administrators:
253 "perlbug-admin" <perlbug-admin@perl.org>
255 =head2 Submitting patches
257 Always submit patches to I<perl5-porters@perl.org>. If you're
258 patching a core module and there's an author listed, send the author a
259 copy (see L<Patching a core module>). This lets other porters review
260 your patch, which catches a surprising number of errors in patches.
261 Please patch against the latest B<development> version. (e.g., even if
262 you're fixing a bug in the 5.8 track, patch against the C<blead> branch in
265 If changes are accepted, they are applied to the development branch. Then
266 the maintenance pumpking decides which of those patches is to be
267 backported to the maint branch. Only patches that survive the heat of the
268 development branch get applied to maintenance versions.
270 Your patch should update the documentation and test suite. See
271 L<Writing a test>. If you have added or removed files in the distribution,
272 edit the MANIFEST file accordingly, sort the MANIFEST file using
273 C<make manisort>, and include those changes as part of your patch.
275 Patching documentation also follows the same order: if accepted, a patch
276 is first applied to B<development>, and if relevant then it's backported
277 to B<maintenance>. (With an exception for some patches that document
278 behaviour that only appears in the maintenance branch, but which has
279 changed in the development version.)
281 To report a bug in Perl, use the program I<perlbug> which comes with
282 Perl (if you can't get Perl to work, send mail to the address
283 I<perlbug@perl.org> or I<perlbug@perl.com>). Reporting bugs through
284 I<perlbug> feeds into the automated bug-tracking system, access to
285 which is provided through the web at http://rt.perl.org/rt3/ . It
286 often pays to check the archives of the perl5-porters mailing list to
287 see whether the bug you're reporting has been reported before, and if
288 so whether it was considered a bug. See above for the location of
289 the searchable archives.
291 The CPAN testers ( http://testers.cpan.org/ ) are a group of
292 volunteers who test CPAN modules on a variety of platforms. Perl
293 Smokers ( http://www.nntp.perl.org/group/perl.daily-build and
294 http://www.nntp.perl.org/group/perl.daily-build.reports/ )
295 automatically test Perl source releases on platforms with various
296 configurations. Both efforts welcome volunteers. In order to get
297 involved in smoke testing of the perl itself visit
298 L<http://search.cpan.org/dist/Test-Smoke>. In order to start smoke
299 testing CPAN modules visit L<http://search.cpan.org/dist/CPAN-YACSmoke/>
300 or L<http://search.cpan.org/dist/POE-Component-CPAN-YACSmoke/> or
301 L<http://search.cpan.org/dist/CPAN-Reporter/>.
303 It's a good idea to read and lurk for a while before chipping in.
304 That way you'll get to see the dynamic of the conversations, learn the
305 personalities of the players, and hopefully be better prepared to make
306 a useful contribution when do you speak up.
308 If after all this you still think you want to join the perl5-porters
309 mailing list, send mail to I<perl5-porters-subscribe@perl.org>. To
310 unsubscribe, send mail to I<perl5-porters-unsubscribe@perl.org>.
312 To hack on the Perl guts, you'll need to read the following things:
318 This is of paramount importance, since it's the documentation of what
319 goes where in the Perl source. Read it over a couple of times and it
320 might start to make sense - don't worry if it doesn't yet, because the
321 best way to study it is to read it in conjunction with poking at Perl
322 source, and we'll do that later on.
324 Gisle Aas's illustrated perlguts (also known as I<illguts>) is wonderful,
325 although a little out of date with regard to some size details; the
326 various SV structures have since been reworked for smaller memory footprint.
327 The fundamentals are right however, and the pictures are very helpful.
329 L<http://www.perl.org/tpc/1998/Perl_Language_and_Modules/Perl%20Illustrated/>
331 =item L<perlxstut> and L<perlxs>
333 A working knowledge of XSUB programming is incredibly useful for core
334 hacking; XSUBs use techniques drawn from the PP code, the portion of the
335 guts that actually executes a Perl program. It's a lot gentler to learn
336 those techniques from simple examples and explanation than from the core
341 The documentation for the Perl API explains what some of the internal
342 functions do, as well as the many macros used in the source.
344 =item F<Porting/pumpkin.pod>
346 This is a collection of words of wisdom for a Perl porter; some of it is
347 only useful to the pumpkin holder, but most of it applies to anyone
348 wanting to go about Perl development.
350 =item The perl5-porters FAQ
352 This should be available from http://dev.perl.org/perl5/docs/p5p-faq.html .
353 It contains hints on reading perl5-porters, information on how
354 perl5-porters works and how Perl development in general works.
358 =head2 Finding Your Way Around
360 Perl maintenance can be split into a number of areas, and certain people
361 (pumpkins) will have responsibility for each area. These areas sometimes
362 correspond to files or directories in the source kit. Among the areas are:
368 Modules shipped as part of the Perl core live in various subdirectories, where
369 two are dedicated to core-only modules, and two are for the dual-life modules
370 which live on CPAN and may be maintained separately with respect to the Perl
373 lib/ is for pure-Perl modules, which exist in the core only.
375 ext/ is for XS extensions, and modules with special Makefile.PL requirements, which exist in the core only.
377 cpan/ is for dual-life modules, where the CPAN module is canonical (should be patched first).
379 dist/ is for dual-life modules, where the blead source is canonical.
383 There are tests for nearly all the modules, built-ins and major bits
384 of functionality. Test files all have a .t suffix. Module tests live
385 in the F<lib/> and F<ext/> directories next to the module being
386 tested. Others live in F<t/>. See L<Writing a test>
390 Documentation maintenance includes looking after everything in the
391 F<pod/> directory, (as well as contributing new documentation) and
392 the documentation to the modules in core.
396 The configure process is the way we make Perl portable across the
397 myriad of operating systems it supports. Responsibility for the
398 configure, build and installation process, as well as the overall
399 portability of the core code rests with the configure pumpkin - others
400 help out with individual operating systems.
402 The files involved are the operating system directories, (F<win32/>,
403 F<os2/>, F<vms/> and so on) the shell scripts which generate F<config.h>
404 and F<Makefile>, as well as the metaconfig files which generate
405 F<Configure>. (metaconfig isn't included in the core distribution.)
409 And of course, there's the core of the Perl interpreter itself. Let's
410 have a look at that in a little more detail.
414 Before we leave looking at the layout, though, don't forget that
415 F<MANIFEST> contains not only the file names in the Perl distribution,
416 but short descriptions of what's in them, too. For an overview of the
417 important files, try this:
419 perl -lne 'print if /^[^\/]+\.[ch]\s+/' MANIFEST
421 =head2 Elements of the interpreter
423 The work of the interpreter has two main stages: compiling the code
424 into the internal representation, or bytecode, and then executing it.
425 L<perlguts/Compiled code> explains exactly how the compilation stage
428 Here is a short breakdown of perl's operation:
434 The action begins in F<perlmain.c>. (or F<miniperlmain.c> for miniperl)
435 This is very high-level code, enough to fit on a single screen, and it
436 resembles the code found in L<perlembed>; most of the real action takes
439 F<perlmain.c> is generated by L<writemain> from F<miniperlmain.c> at
440 make time, so you should make perl to follow this along.
442 First, F<perlmain.c> allocates some memory and constructs a Perl
443 interpreter, along these lines:
445 1 PERL_SYS_INIT3(&argc,&argv,&env);
447 3 if (!PL_do_undump) {
448 4 my_perl = perl_alloc();
451 7 perl_construct(my_perl);
452 8 PL_perl_destruct_level = 0;
455 Line 1 is a macro, and its definition is dependent on your operating
456 system. Line 3 references C<PL_do_undump>, a global variable - all
457 global variables in Perl start with C<PL_>. This tells you whether the
458 current running program was created with the C<-u> flag to perl and then
459 F<undump>, which means it's going to be false in any sane context.
461 Line 4 calls a function in F<perl.c> to allocate memory for a Perl
462 interpreter. It's quite a simple function, and the guts of it looks like
465 my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
467 Here you see an example of Perl's system abstraction, which we'll see
468 later: C<PerlMem_malloc> is either your system's C<malloc>, or Perl's
469 own C<malloc> as defined in F<malloc.c> if you selected that option at
472 Next, in line 7, we construct the interpreter using perl_construct,
473 also in F<perl.c>; this sets up all the special variables that Perl
474 needs, the stacks, and so on.
476 Now we pass Perl the command line options, and tell it to go:
478 exitstatus = perl_parse(my_perl, xs_init, argc, argv, (char **)NULL);
482 exitstatus = perl_destruct(my_perl);
486 C<perl_parse> is actually a wrapper around C<S_parse_body>, as defined
487 in F<perl.c>, which processes the command line options, sets up any
488 statically linked XS modules, opens the program and calls C<yyparse> to
493 The aim of this stage is to take the Perl source, and turn it into an op
494 tree. We'll see what one of those looks like later. Strictly speaking,
495 there's three things going on here.
497 C<yyparse>, the parser, lives in F<perly.c>, although you're better off
498 reading the original YACC input in F<perly.y>. (Yes, Virginia, there
499 B<is> a YACC grammar for Perl!) The job of the parser is to take your
500 code and "understand" it, splitting it into sentences, deciding which
501 operands go with which operators and so on.
503 The parser is nobly assisted by the lexer, which chunks up your input
504 into tokens, and decides what type of thing each token is: a variable
505 name, an operator, a bareword, a subroutine, a core function, and so on.
506 The main point of entry to the lexer is C<yylex>, and that and its
507 associated routines can be found in F<toke.c>. Perl isn't much like
508 other computer languages; it's highly context sensitive at times, it can
509 be tricky to work out what sort of token something is, or where a token
510 ends. As such, there's a lot of interplay between the tokeniser and the
511 parser, which can get pretty frightening if you're not used to it.
513 As the parser understands a Perl program, it builds up a tree of
514 operations for the interpreter to perform during execution. The routines
515 which construct and link together the various operations are to be found
516 in F<op.c>, and will be examined later.
520 Now the parsing stage is complete, and the finished tree represents
521 the operations that the Perl interpreter needs to perform to execute our
522 program. Next, Perl does a dry run over the tree looking for
523 optimisations: constant expressions such as C<3 + 4> will be computed
524 now, and the optimizer will also see if any multiple operations can be
525 replaced with a single one. For instance, to fetch the variable C<$foo>,
526 instead of grabbing the glob C<*foo> and looking at the scalar
527 component, the optimizer fiddles the op tree to use a function which
528 directly looks up the scalar in question. The main optimizer is C<peep>
529 in F<op.c>, and many ops have their own optimizing functions.
533 Now we're finally ready to go: we have compiled Perl byte code, and all
534 that's left to do is run it. The actual execution is done by the
535 C<runops_standard> function in F<run.c>; more specifically, it's done by
536 these three innocent looking lines:
538 while ((PL_op = CALL_FPTR(PL_op->op_ppaddr)(aTHX))) {
542 You may be more comfortable with the Perl version of that:
544 PERL_ASYNC_CHECK() while $Perl::op = &{$Perl::op->{function}};
546 Well, maybe not. Anyway, each op contains a function pointer, which
547 stipulates the function which will actually carry out the operation.
548 This function will return the next op in the sequence - this allows for
549 things like C<if> which choose the next op dynamically at run time.
550 The C<PERL_ASYNC_CHECK> makes sure that things like signals interrupt
551 execution if required.
553 The actual functions called are known as PP code, and they're spread
554 between four files: F<pp_hot.c> contains the "hot" code, which is most
555 often used and highly optimized, F<pp_sys.c> contains all the
556 system-specific functions, F<pp_ctl.c> contains the functions which
557 implement control structures (C<if>, C<while> and the like) and F<pp.c>
558 contains everything else. These are, if you like, the C code for Perl's
559 built-in functions and operators.
561 Note that each C<pp_> function is expected to return a pointer to the next
562 op. Calls to perl subs (and eval blocks) are handled within the same
563 runops loop, and do not consume extra space on the C stack. For example,
564 C<pp_entersub> and C<pp_entertry> just push a C<CxSUB> or C<CxEVAL> block
565 struct onto the context stack which contain the address of the op
566 following the sub call or eval. They then return the first op of that sub
567 or eval block, and so execution continues of that sub or block. Later, a
568 C<pp_leavesub> or C<pp_leavetry> op pops the C<CxSUB> or C<CxEVAL>,
569 retrieves the return op from it, and returns it.
571 =item Exception handing
573 Perl's exception handing (i.e. C<die> etc.) is built on top of the low-level
574 C<setjmp()>/C<longjmp()> C-library functions. These basically provide a
575 way to capture the current PC and SP registers and later restore them; i.e.
576 a C<longjmp()> continues at the point in code where a previous C<setjmp()>
577 was done, with anything further up on the C stack being lost. This is why
578 code should always save values using C<SAVE_FOO> rather than in auto
581 The perl core wraps C<setjmp()> etc in the macros C<JMPENV_PUSH> and
582 C<JMPENV_JUMP>. The basic rule of perl exceptions is that C<exit>, and
583 C<die> (in the absence of C<eval>) perform a C<JMPENV_JUMP(2)>, while
584 C<die> within C<eval> does a C<JMPENV_JUMP(3)>.
586 At entry points to perl, such as C<perl_parse()>, C<perl_run()> and
587 C<call_sv(cv, G_EVAL)> each does a C<JMPENV_PUSH>, then enter a runops
588 loop or whatever, and handle possible exception returns. For a 2 return,
589 final cleanup is performed, such as popping stacks and calling C<CHECK> or
590 C<END> blocks. Amongst other things, this is how scope cleanup still
591 occurs during an C<exit>.
593 If a C<die> can find a C<CxEVAL> block on the context stack, then the
594 stack is popped to that level and the return op in that block is assigned
595 to C<PL_restartop>; then a C<JMPENV_JUMP(3)> is performed. This normally
596 passes control back to the guard. In the case of C<perl_run> and
597 C<call_sv>, a non-null C<PL_restartop> triggers re-entry to the runops
598 loop. The is the normal way that C<die> or C<croak> is handled within an
601 Sometimes ops are executed within an inner runops loop, such as tie, sort
602 or overload code. In this case, something like
604 sub FETCH { eval { die } }
606 would cause a longjmp right back to the guard in C<perl_run>, popping both
607 runops loops, which is clearly incorrect. One way to avoid this is for the
608 tie code to do a C<JMPENV_PUSH> before executing C<FETCH> in the inner
609 runops loop, but for efficiency reasons, perl in fact just sets a flag,
610 using C<CATCH_SET(TRUE)>. The C<pp_require>, C<pp_entereval> and
611 C<pp_entertry> ops check this flag, and if true, they call C<docatch>,
612 which does a C<JMPENV_PUSH> and starts a new runops level to execute the
613 code, rather than doing it on the current loop.
615 As a further optimisation, on exit from the eval block in the C<FETCH>,
616 execution of the code following the block is still carried on in the inner
617 loop. When an exception is raised, C<docatch> compares the C<JMPENV>
618 level of the C<CxEVAL> with C<PL_top_env> and if they differ, just
619 re-throws the exception. In this way any inner loops get popped.
623 1: eval { tie @a, 'A' };
629 To run this code, C<perl_run> is called, which does a C<JMPENV_PUSH> then
630 enters a runops loop. This loop executes the eval and tie ops on line 1,
631 with the eval pushing a C<CxEVAL> onto the context stack.
633 The C<pp_tie> does a C<CATCH_SET(TRUE)>, then starts a second runops loop
634 to execute the body of C<TIEARRAY>. When it executes the entertry op on
635 line 3, C<CATCH_GET> is true, so C<pp_entertry> calls C<docatch> which
636 does a C<JMPENV_PUSH> and starts a third runops loop, which then executes
637 the die op. At this point the C call stack looks like this:
640 Perl_runops # third loop
644 Perl_runops # second loop
648 Perl_runops # first loop
653 and the context and data stacks, as shown by C<-Dstv>, look like:
657 CX 1: EVAL => AV() PV("A"\0)
665 The die pops the first C<CxEVAL> off the context stack, sets
666 C<PL_restartop> from it, does a C<JMPENV_JUMP(3)>, and control returns to
667 the top C<docatch>. This then starts another third-level runops level,
668 which executes the nextstate, pushmark and die ops on line 4. At the point
669 that the second C<pp_die> is called, the C call stack looks exactly like
670 that above, even though we are no longer within an inner eval; this is
671 because of the optimization mentioned earlier. However, the context stack
672 now looks like this, ie with the top CxEVAL popped:
676 CX 1: EVAL => AV() PV("A"\0)
682 The die on line 4 pops the context stack back down to the CxEVAL, leaving
688 As usual, C<PL_restartop> is extracted from the C<CxEVAL>, and a
689 C<JMPENV_JUMP(3)> done, which pops the C stack back to the docatch:
693 Perl_runops # second loop
697 Perl_runops # first loop
702 In this case, because the C<JMPENV> level recorded in the C<CxEVAL>
703 differs from the current one, C<docatch> just does a C<JMPENV_JUMP(3)>
704 and the C stack unwinds to:
709 Because C<PL_restartop> is non-null, C<run_body> starts a new runops loop
710 and execution continues.
714 =head2 Internal Variable Types
716 You should by now have had a look at L<perlguts>, which tells you about
717 Perl's internal variable types: SVs, HVs, AVs and the rest. If not, do
720 These variables are used not only to represent Perl-space variables, but
721 also any constants in the code, as well as some structures completely
722 internal to Perl. The symbol table, for instance, is an ordinary Perl
723 hash. Your code is represented by an SV as it's read into the parser;
724 any program files you call are opened via ordinary Perl filehandles, and
727 The core L<Devel::Peek|Devel::Peek> module lets us examine SVs from a
728 Perl program. Let's see, for instance, how Perl treats the constant
731 % perl -MDevel::Peek -e 'Dump("hello")'
732 1 SV = PV(0xa041450) at 0xa04ecbc
734 3 FLAGS = (POK,READONLY,pPOK)
735 4 PV = 0xa0484e0 "hello"\0
739 Reading C<Devel::Peek> output takes a bit of practise, so let's go
740 through it line by line.
742 Line 1 tells us we're looking at an SV which lives at C<0xa04ecbc> in
743 memory. SVs themselves are very simple structures, but they contain a
744 pointer to a more complex structure. In this case, it's a PV, a
745 structure which holds a string value, at location C<0xa041450>. Line 2
746 is the reference count; there are no other references to this data, so
749 Line 3 are the flags for this SV - it's OK to use it as a PV, it's a
750 read-only SV (because it's a constant) and the data is a PV internally.
751 Next we've got the contents of the string, starting at location
754 Line 5 gives us the current length of the string - note that this does
755 B<not> include the null terminator. Line 6 is not the length of the
756 string, but the length of the currently allocated buffer; as the string
757 grows, Perl automatically extends the available storage via a routine
760 You can get at any of these quantities from C very easily; just add
761 C<Sv> to the name of the field shown in the snippet, and you've got a
762 macro which will return the value: C<SvCUR(sv)> returns the current
763 length of the string, C<SvREFCOUNT(sv)> returns the reference count,
764 C<SvPV(sv, len)> returns the string itself with its length, and so on.
765 More macros to manipulate these properties can be found in L<perlguts>.
767 Let's take an example of manipulating a PV, from C<sv_catpvn>, in F<sv.c>
770 2 Perl_sv_catpvn(pTHX_ register SV *sv, register const char *ptr, register STRLEN len)
775 6 junk = SvPV_force(sv, tlen);
776 7 SvGROW(sv, tlen + len + 1);
779 10 Move(ptr,SvPVX(sv)+tlen,len,char);
781 12 *SvEND(sv) = '\0';
782 13 (void)SvPOK_only_UTF8(sv); /* validate pointer */
786 This is a function which adds a string, C<ptr>, of length C<len> onto
787 the end of the PV stored in C<sv>. The first thing we do in line 6 is
788 make sure that the SV B<has> a valid PV, by calling the C<SvPV_force>
789 macro to force a PV. As a side effect, C<tlen> gets set to the current
790 value of the PV, and the PV itself is returned to C<junk>.
792 In line 7, we make sure that the SV will have enough room to accommodate
793 the old string, the new string and the null terminator. If C<LEN> isn't
794 big enough, C<SvGROW> will reallocate space for us.
796 Now, if C<junk> is the same as the string we're trying to add, we can
797 grab the string directly from the SV; C<SvPVX> is the address of the PV
800 Line 10 does the actual catenation: the C<Move> macro moves a chunk of
801 memory around: we move the string C<ptr> to the end of the PV - that's
802 the start of the PV plus its current length. We're moving C<len> bytes
803 of type C<char>. After doing so, we need to tell Perl we've extended the
804 string, by altering C<CUR> to reflect the new length. C<SvEND> is a
805 macro which gives us the end of the string, so that needs to be a
808 Line 13 manipulates the flags; since we've changed the PV, any IV or NV
809 values will no longer be valid: if we have C<$a=10; $a.="6";> we don't
810 want to use the old IV of 10. C<SvPOK_only_utf8> is a special UTF-8-aware
811 version of C<SvPOK_only>, a macro which turns off the IOK and NOK flags
812 and turns on POK. The final C<SvTAINT> is a macro which launders tainted
813 data if taint mode is turned on.
815 AVs and HVs are more complicated, but SVs are by far the most common
816 variable type being thrown around. Having seen something of how we
817 manipulate these, let's go on and look at how the op tree is
822 First, what is the op tree, anyway? The op tree is the parsed
823 representation of your program, as we saw in our section on parsing, and
824 it's the sequence of operations that Perl goes through to execute your
825 program, as we saw in L</Running>.
827 An op is a fundamental operation that Perl can perform: all the built-in
828 functions and operators are ops, and there are a series of ops which
829 deal with concepts the interpreter needs internally - entering and
830 leaving a block, ending a statement, fetching a variable, and so on.
832 The op tree is connected in two ways: you can imagine that there are two
833 "routes" through it, two orders in which you can traverse the tree.
834 First, parse order reflects how the parser understood the code, and
835 secondly, execution order tells perl what order to perform the
838 The easiest way to examine the op tree is to stop Perl after it has
839 finished parsing, and get it to dump out the tree. This is exactly what
840 the compiler backends L<B::Terse|B::Terse>, L<B::Concise|B::Concise>
841 and L<B::Debug|B::Debug> do.
843 Let's have a look at how Perl sees C<$a = $b + $c>:
845 % perl -MO=Terse -e '$a=$b+$c'
846 1 LISTOP (0x8179888) leave
847 2 OP (0x81798b0) enter
848 3 COP (0x8179850) nextstate
849 4 BINOP (0x8179828) sassign
850 5 BINOP (0x8179800) add [1]
851 6 UNOP (0x81796e0) null [15]
852 7 SVOP (0x80fafe0) gvsv GV (0x80fa4cc) *b
853 8 UNOP (0x81797e0) null [15]
854 9 SVOP (0x8179700) gvsv GV (0x80efeb0) *c
855 10 UNOP (0x816b4f0) null [15]
856 11 SVOP (0x816dcf0) gvsv GV (0x80fa460) *a
858 Let's start in the middle, at line 4. This is a BINOP, a binary
859 operator, which is at location C<0x8179828>. The specific operator in
860 question is C<sassign> - scalar assignment - and you can find the code
861 which implements it in the function C<pp_sassign> in F<pp_hot.c>. As a
862 binary operator, it has two children: the add operator, providing the
863 result of C<$b+$c>, is uppermost on line 5, and the left hand side is on
866 Line 10 is the null op: this does exactly nothing. What is that doing
867 there? If you see the null op, it's a sign that something has been
868 optimized away after parsing. As we mentioned in L</Optimization>,
869 the optimization stage sometimes converts two operations into one, for
870 example when fetching a scalar variable. When this happens, instead of
871 rewriting the op tree and cleaning up the dangling pointers, it's easier
872 just to replace the redundant operation with the null op. Originally,
873 the tree would have looked like this:
875 10 SVOP (0x816b4f0) rv2sv [15]
876 11 SVOP (0x816dcf0) gv GV (0x80fa460) *a
878 That is, fetch the C<a> entry from the main symbol table, and then look
879 at the scalar component of it: C<gvsv> (C<pp_gvsv> into F<pp_hot.c>)
880 happens to do both these things.
882 The right hand side, starting at line 5 is similar to what we've just
883 seen: we have the C<add> op (C<pp_add> also in F<pp_hot.c>) add together
886 Now, what's this about?
888 1 LISTOP (0x8179888) leave
889 2 OP (0x81798b0) enter
890 3 COP (0x8179850) nextstate
892 C<enter> and C<leave> are scoping ops, and their job is to perform any
893 housekeeping every time you enter and leave a block: lexical variables
894 are tidied up, unreferenced variables are destroyed, and so on. Every
895 program will have those first three lines: C<leave> is a list, and its
896 children are all the statements in the block. Statements are delimited
897 by C<nextstate>, so a block is a collection of C<nextstate> ops, with
898 the ops to be performed for each statement being the children of
899 C<nextstate>. C<enter> is a single op which functions as a marker.
901 That's how Perl parsed the program, from top to bottom:
914 However, it's impossible to B<perform> the operations in this order:
915 you have to find the values of C<$b> and C<$c> before you add them
916 together, for instance. So, the other thread that runs through the op
917 tree is the execution order: each op has a field C<op_next> which points
918 to the next op to be run, so following these pointers tells us how perl
919 executes the code. We can traverse the tree in this order using
920 the C<exec> option to C<B::Terse>:
922 % perl -MO=Terse,exec -e '$a=$b+$c'
923 1 OP (0x8179928) enter
924 2 COP (0x81798c8) nextstate
925 3 SVOP (0x81796c8) gvsv GV (0x80fa4d4) *b
926 4 SVOP (0x8179798) gvsv GV (0x80efeb0) *c
927 5 BINOP (0x8179878) add [1]
928 6 SVOP (0x816dd38) gvsv GV (0x80fa468) *a
929 7 BINOP (0x81798a0) sassign
930 8 LISTOP (0x8179900) leave
932 This probably makes more sense for a human: enter a block, start a
933 statement. Get the values of C<$b> and C<$c>, and add them together.
934 Find C<$a>, and assign one to the other. Then leave.
936 The way Perl builds up these op trees in the parsing process can be
937 unravelled by examining F<perly.y>, the YACC grammar. Let's take the
938 piece we need to construct the tree for C<$a = $b + $c>
940 1 term : term ASSIGNOP term
941 2 { $$ = newASSIGNOP(OPf_STACKED, $1, $2, $3); }
943 4 { $$ = newBINOP($2, 0, scalar($1), scalar($3)); }
945 If you're not used to reading BNF grammars, this is how it works: You're
946 fed certain things by the tokeniser, which generally end up in upper
947 case. Here, C<ADDOP>, is provided when the tokeniser sees C<+> in your
948 code. C<ASSIGNOP> is provided when C<=> is used for assigning. These are
949 "terminal symbols", because you can't get any simpler than them.
951 The grammar, lines one and three of the snippet above, tells you how to
952 build up more complex forms. These complex forms, "non-terminal symbols"
953 are generally placed in lower case. C<term> here is a non-terminal
954 symbol, representing a single expression.
956 The grammar gives you the following rule: you can make the thing on the
957 left of the colon if you see all the things on the right in sequence.
958 This is called a "reduction", and the aim of parsing is to completely
959 reduce the input. There are several different ways you can perform a
960 reduction, separated by vertical bars: so, C<term> followed by C<=>
961 followed by C<term> makes a C<term>, and C<term> followed by C<+>
962 followed by C<term> can also make a C<term>.
964 So, if you see two terms with an C<=> or C<+>, between them, you can
965 turn them into a single expression. When you do this, you execute the
966 code in the block on the next line: if you see C<=>, you'll do the code
967 in line 2. If you see C<+>, you'll do the code in line 4. It's this code
968 which contributes to the op tree.
971 { $$ = newBINOP($2, 0, scalar($1), scalar($3)); }
973 What this does is creates a new binary op, and feeds it a number of
974 variables. The variables refer to the tokens: C<$1> is the first token in
975 the input, C<$2> the second, and so on - think regular expression
976 backreferences. C<$$> is the op returned from this reduction. So, we
977 call C<newBINOP> to create a new binary operator. The first parameter to
978 C<newBINOP>, a function in F<op.c>, is the op type. It's an addition
979 operator, so we want the type to be C<ADDOP>. We could specify this
980 directly, but it's right there as the second token in the input, so we
981 use C<$2>. The second parameter is the op's flags: 0 means "nothing
982 special". Then the things to add: the left and right hand side of our
983 expression, in scalar context.
987 When perl executes something like C<addop>, how does it pass on its
988 results to the next op? The answer is, through the use of stacks. Perl
989 has a number of stacks to store things it's currently working on, and
990 we'll look at the three most important ones here.
996 Arguments are passed to PP code and returned from PP code using the
997 argument stack, C<ST>. The typical way to handle arguments is to pop
998 them off the stack, deal with them how you wish, and then push the result
999 back onto the stack. This is how, for instance, the cosine operator
1004 value = Perl_cos(value);
1007 We'll see a more tricky example of this when we consider Perl's macros
1008 below. C<POPn> gives you the NV (floating point value) of the top SV on
1009 the stack: the C<$x> in C<cos($x)>. Then we compute the cosine, and push
1010 the result back as an NV. The C<X> in C<XPUSHn> means that the stack
1011 should be extended if necessary - it can't be necessary here, because we
1012 know there's room for one more item on the stack, since we've just
1013 removed one! The C<XPUSH*> macros at least guarantee safety.
1015 Alternatively, you can fiddle with the stack directly: C<SP> gives you
1016 the first element in your portion of the stack, and C<TOP*> gives you
1017 the top SV/IV/NV/etc. on the stack. So, for instance, to do unary
1018 negation of an integer:
1022 Just set the integer value of the top stack entry to its negation.
1024 Argument stack manipulation in the core is exactly the same as it is in
1025 XSUBs - see L<perlxstut>, L<perlxs> and L<perlguts> for a longer
1026 description of the macros used in stack manipulation.
1030 I say "your portion of the stack" above because PP code doesn't
1031 necessarily get the whole stack to itself: if your function calls
1032 another function, you'll only want to expose the arguments aimed for the
1033 called function, and not (necessarily) let it get at your own data. The
1034 way we do this is to have a "virtual" bottom-of-stack, exposed to each
1035 function. The mark stack keeps bookmarks to locations in the argument
1036 stack usable by each function. For instance, when dealing with a tied
1037 variable, (internally, something with "P" magic) Perl has to call
1038 methods for accesses to the tied variables. However, we need to separate
1039 the arguments exposed to the method to the argument exposed to the
1040 original function - the store or fetch or whatever it may be. Here's
1041 roughly how the tied C<push> is implemented; see C<av_push> in F<av.c>:
1045 3 PUSHs(SvTIED_obj((SV*)av, mg));
1049 7 call_method("PUSH", G_SCALAR|G_DISCARD);
1052 Let's examine the whole implementation, for practice:
1056 Push the current state of the stack pointer onto the mark stack. This is
1057 so that when we've finished adding items to the argument stack, Perl
1058 knows how many things we've added recently.
1061 3 PUSHs(SvTIED_obj((SV*)av, mg));
1064 We're going to add two more items onto the argument stack: when you have
1065 a tied array, the C<PUSH> subroutine receives the object and the value
1066 to be pushed, and that's exactly what we have here - the tied object,
1067 retrieved with C<SvTIED_obj>, and the value, the SV C<val>.
1071 Next we tell Perl to update the global stack pointer from our internal
1072 variable: C<dSP> only gave us a local copy, not a reference to the global.
1075 7 call_method("PUSH", G_SCALAR|G_DISCARD);
1078 C<ENTER> and C<LEAVE> localise a block of code - they make sure that all
1079 variables are tidied up, everything that has been localised gets
1080 its previous value returned, and so on. Think of them as the C<{> and
1081 C<}> of a Perl block.
1083 To actually do the magic method call, we have to call a subroutine in
1084 Perl space: C<call_method> takes care of that, and it's described in
1085 L<perlcall>. We call the C<PUSH> method in scalar context, and we're
1086 going to discard its return value. The call_method() function
1087 removes the top element of the mark stack, so there is nothing for
1088 the caller to clean up.
1092 C doesn't have a concept of local scope, so perl provides one. We've
1093 seen that C<ENTER> and C<LEAVE> are used as scoping braces; the save
1094 stack implements the C equivalent of, for example:
1101 See L<perlguts/Localising Changes> for how to use the save stack.
1105 =head2 Millions of Macros
1107 One thing you'll notice about the Perl source is that it's full of
1108 macros. Some have called the pervasive use of macros the hardest thing
1109 to understand, others find it adds to clarity. Let's take an example,
1110 the code which implements the addition operator:
1114 3 dSP; dATARGET; tryAMAGICbin(add,opASSIGN);
1117 6 SETn( left + right );
1122 Every line here (apart from the braces, of course) contains a macro. The
1123 first line sets up the function declaration as Perl expects for PP code;
1124 line 3 sets up variable declarations for the argument stack and the
1125 target, the return value of the operation. Finally, it tries to see if
1126 the addition operation is overloaded; if so, the appropriate subroutine
1129 Line 5 is another variable declaration - all variable declarations start
1130 with C<d> - which pops from the top of the argument stack two NVs (hence
1131 C<nn>) and puts them into the variables C<right> and C<left>, hence the
1132 C<rl>. These are the two operands to the addition operator. Next, we
1133 call C<SETn> to set the NV of the return value to the result of adding
1134 the two values. This done, we return - the C<RETURN> macro makes sure
1135 that our return value is properly handled, and we pass the next operator
1136 to run back to the main run loop.
1138 Most of these macros are explained in L<perlapi>, and some of the more
1139 important ones are explained in L<perlxs> as well. Pay special attention
1140 to L<perlguts/Background and PERL_IMPLICIT_CONTEXT> for information on
1141 the C<[pad]THX_?> macros.
1143 =head2 The .i Targets
1145 You can expand the macros in a F<foo.c> file by saying
1149 which will expand the macros using cpp. Don't be scared by the results.
1151 =head1 SOURCE CODE STATIC ANALYSIS
1153 Various tools exist for analysing C source code B<statically>, as
1154 opposed to B<dynamically>, that is, without executing the code.
1155 It is possible to detect resource leaks, undefined behaviour, type
1156 mismatches, portability problems, code paths that would cause illegal
1157 memory accesses, and other similar problems by just parsing the C code
1158 and looking at the resulting graph, what does it tell about the
1159 execution and data flows. As a matter of fact, this is exactly
1160 how C compilers know to give warnings about dubious code.
1164 The good old C code quality inspector, C<lint>, is available in
1165 several platforms, but please be aware that there are several
1166 different implementations of it by different vendors, which means that
1167 the flags are not identical across different platforms.
1169 There is a lint variant called C<splint> (Secure Programming Lint)
1170 available from http://www.splint.org/ that should compile on any
1173 There are C<lint> and <splint> targets in Makefile, but you may have
1174 to diddle with the flags (see above).
1178 Coverity (http://www.coverity.com/) is a product similar to lint and
1179 as a testbed for their product they periodically check several open
1180 source projects, and they give out accounts to open source developers
1181 to the defect databases.
1183 =head2 cpd (cut-and-paste detector)
1185 The cpd tool detects cut-and-paste coding. If one instance of the
1186 cut-and-pasted code changes, all the other spots should probably be
1187 changed, too. Therefore such code should probably be turned into a
1188 subroutine or a macro.
1190 cpd (http://pmd.sourceforge.net/cpd.html) is part of the pmd project
1191 (http://pmd.sourceforge.net/). pmd was originally written for static
1192 analysis of Java code, but later the cpd part of it was extended to
1193 parse also C and C++.
1195 Download the pmd-bin-X.Y.zip () from the SourceForge site, extract the
1196 pmd-X.Y.jar from it, and then run that on source code thusly:
1198 java -cp pmd-X.Y.jar net.sourceforge.pmd.cpd.CPD --minimum-tokens 100 --files /some/where/src --language c > cpd.txt
1200 You may run into memory limits, in which case you should use the -Xmx option:
1206 Though much can be written about the inconsistency and coverage
1207 problems of gcc warnings (like C<-Wall> not meaning "all the
1208 warnings", or some common portability problems not being covered by
1209 C<-Wall>, or C<-ansi> and C<-pedantic> both being a poorly defined
1210 collection of warnings, and so forth), gcc is still a useful tool in
1211 keeping our coding nose clean.
1213 The C<-Wall> is by default on.
1215 The C<-ansi> (and its sidekick, C<-pedantic>) would be nice to be on
1216 always, but unfortunately they are not safe on all platforms, they can
1217 for example cause fatal conflicts with the system headers (Solaris
1218 being a prime example). If Configure C<-Dgccansipedantic> is used,
1219 the C<cflags> frontend selects C<-ansi -pedantic> for the platforms
1220 where they are known to be safe.
1222 Starting from Perl 5.9.4 the following extra flags are added:
1236 C<-Wdeclaration-after-statement>
1240 The following flags would be nice to have but they would first need
1241 their own Augean stablemaster:
1255 C<-Wstrict-prototypes>
1259 The C<-Wtraditional> is another example of the annoying tendency of
1260 gcc to bundle a lot of warnings under one switch -- it would be
1261 impossible to deploy in practice because it would complain a lot -- but
1262 it does contain some warnings that would be beneficial to have available
1263 on their own, such as the warning about string constants inside macros
1264 containing the macro arguments: this behaved differently pre-ANSI
1265 than it does in ANSI, and some C compilers are still in transition,
1266 AIX being an example.
1268 =head2 Warnings of other C compilers
1270 Other C compilers (yes, there B<are> other C compilers than gcc) often
1271 have their "strict ANSI" or "strict ANSI with some portability extensions"
1272 modes on, like for example the Sun Workshop has its C<-Xa> mode on
1273 (though implicitly), or the DEC (these days, HP...) has its C<-std1>
1278 You can compile a special debugging version of Perl, which allows you
1279 to use the C<-D> option of Perl to tell more about what Perl is doing.
1280 But sometimes there is no alternative than to dive in with a debugger,
1281 either to see the stack trace of a core dump (very useful in a bug
1282 report), or trying to figure out what went wrong before the core dump
1283 happened, or how did we end up having wrong or unexpected results.
1285 =head2 Poking at Perl
1287 To really poke around with Perl, you'll probably want to build Perl for
1288 debugging, like this:
1290 ./Configure -d -D optimize=-g
1293 C<-g> is a flag to the C compiler to have it produce debugging
1294 information which will allow us to step through a running program,
1295 and to see in which C function we are at (without the debugging
1296 information we might see only the numerical addresses of the functions,
1297 which is not very helpful).
1299 F<Configure> will also turn on the C<DEBUGGING> compilation symbol which
1300 enables all the internal debugging code in Perl. There are a whole bunch
1301 of things you can debug with this: L<perlrun> lists them all, and the
1302 best way to find out about them is to play about with them. The most
1303 useful options are probably
1305 l Context (loop) stack processing
1307 o Method and overloading resolution
1308 c String/numeric conversions
1310 Some of the functionality of the debugging code can be achieved using XS
1313 -Dr => use re 'debug'
1314 -Dx => use O 'Debug'
1316 =head2 Using a source-level debugger
1318 If the debugging output of C<-D> doesn't help you, it's time to step
1319 through perl's execution with a source-level debugger.
1325 We'll use C<gdb> for our examples here; the principles will apply to
1326 any debugger (many vendors call their debugger C<dbx>), but check the
1327 manual of the one you're using.
1331 To fire up the debugger, type
1335 Or if you have a core dump:
1339 You'll want to do that in your Perl source tree so the debugger can read
1340 the source code. You should see the copyright message, followed by the
1345 C<help> will get you into the documentation, but here are the most
1352 Run the program with the given arguments.
1354 =item break function_name
1356 =item break source.c:xxx
1358 Tells the debugger that we'll want to pause execution when we reach
1359 either the named function (but see L<perlguts/Internal Functions>!) or the given
1360 line in the named source file.
1364 Steps through the program a line at a time.
1368 Steps through the program a line at a time, without descending into
1373 Run until the next breakpoint.
1377 Run until the end of the current function, then stop again.
1381 Just pressing Enter will do the most recent operation again - it's a
1382 blessing when stepping through miles of source code.
1386 Execute the given C code and print its results. B<WARNING>: Perl makes
1387 heavy use of macros, and F<gdb> does not necessarily support macros
1388 (see later L</"gdb macro support">). You'll have to substitute them
1389 yourself, or to invoke cpp on the source code files
1390 (see L</"The .i Targets">)
1391 So, for instance, you can't say
1393 print SvPV_nolen(sv)
1397 print Perl_sv_2pv_nolen(sv)
1401 You may find it helpful to have a "macro dictionary", which you can
1402 produce by saying C<cpp -dM perl.c | sort>. Even then, F<cpp> won't
1403 recursively apply those macros for you.
1405 =head2 gdb macro support
1407 Recent versions of F<gdb> have fairly good macro support, but
1408 in order to use it you'll need to compile perl with macro definitions
1409 included in the debugging information. Using F<gcc> version 3.1, this
1410 means configuring with C<-Doptimize=-g3>. Other compilers might use a
1411 different switch (if they support debugging macros at all).
1413 =head2 Dumping Perl Data Structures
1415 One way to get around this macro hell is to use the dumping functions in
1416 F<dump.c>; these work a little like an internal
1417 L<Devel::Peek|Devel::Peek>, but they also cover OPs and other structures
1418 that you can't get at from Perl. Let's take an example. We'll use the
1419 C<$a = $b + $c> we used before, but give it a bit of context:
1420 C<$b = "6XXXX"; $c = 2.3;>. Where's a good place to stop and poke around?
1422 What about C<pp_add>, the function we examined earlier to implement the
1425 (gdb) break Perl_pp_add
1426 Breakpoint 1 at 0x46249f: file pp_hot.c, line 309.
1428 Notice we use C<Perl_pp_add> and not C<pp_add> - see L<perlguts/Internal Functions>.
1429 With the breakpoint in place, we can run our program:
1431 (gdb) run -e '$b = "6XXXX"; $c = 2.3; $a = $b + $c'
1433 Lots of junk will go past as gdb reads in the relevant source files and
1434 libraries, and then:
1436 Breakpoint 1, Perl_pp_add () at pp_hot.c:309
1437 309 dSP; dATARGET; tryAMAGICbin(add,opASSIGN);
1442 We looked at this bit of code before, and we said that C<dPOPTOPnnrl_ul>
1443 arranges for two C<NV>s to be placed into C<left> and C<right> - let's
1446 #define dPOPTOPnnrl_ul NV right = POPn; \
1447 SV *leftsv = TOPs; \
1448 NV left = USE_LEFT(leftsv) ? SvNV(leftsv) : 0.0
1450 C<POPn> takes the SV from the top of the stack and obtains its NV either
1451 directly (if C<SvNOK> is set) or by calling the C<sv_2nv> function.
1452 C<TOPs> takes the next SV from the top of the stack - yes, C<POPn> uses
1453 C<TOPs> - but doesn't remove it. We then use C<SvNV> to get the NV from
1454 C<leftsv> in the same way as before - yes, C<POPn> uses C<SvNV>.
1456 Since we don't have an NV for C<$b>, we'll have to use C<sv_2nv> to
1457 convert it. If we step again, we'll find ourselves there:
1459 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1669
1463 We can now use C<Perl_sv_dump> to investigate the SV:
1465 SV = PV(0xa057cc0) at 0xa0675d0
1468 PV = 0xa06a510 "6XXXX"\0
1473 We know we're going to get C<6> from this, so let's finish the
1477 Run till exit from #0 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1671
1478 0x462669 in Perl_pp_add () at pp_hot.c:311
1481 We can also dump out this op: the current op is always stored in
1482 C<PL_op>, and we can dump it with C<Perl_op_dump>. This'll give us
1483 similar output to L<B::Debug|B::Debug>.
1486 13 TYPE = add ===> 14
1488 FLAGS = (SCALAR,KIDS)
1490 TYPE = null ===> (12)
1492 FLAGS = (SCALAR,KIDS)
1494 11 TYPE = gvsv ===> 12
1500 # finish this later #
1504 All right, we've now had a look at how to navigate the Perl sources and
1505 some things you'll need to know when fiddling with them. Let's now get
1506 on and create a simple patch. Here's something Larry suggested: if a
1507 C<U> is the first active format during a C<pack>, (for example,
1508 C<pack "U3C8", @stuff>) then the resulting string should be treated as
1511 How do we prepare to fix this up? First we locate the code in question -
1512 the C<pack> happens at runtime, so it's going to be in one of the F<pp>
1513 files. Sure enough, C<pp_pack> is in F<pp.c>. Since we're going to be
1514 altering this file, let's copy it to F<pp.c~>.
1516 [Well, it was in F<pp.c> when this tutorial was written. It has now been
1517 split off with C<pp_unpack> to its own file, F<pp_pack.c>]
1519 Now let's look over C<pp_pack>: we take a pattern into C<pat>, and then
1520 loop over the pattern, taking each format character in turn into
1521 C<datum_type>. Then for each possible format character, we swallow up
1522 the other arguments in the pattern (a field width, an asterisk, and so
1523 on) and convert the next chunk input into the specified format, adding
1524 it onto the output SV C<cat>.
1526 How do we know if the C<U> is the first format in the C<pat>? Well, if
1527 we have a pointer to the start of C<pat> then, if we see a C<U> we can
1528 test whether we're still at the start of the string. So, here's where
1532 register char *pat = SvPVx(*++MARK, fromlen);
1533 register char *patend = pat + fromlen;
1538 We'll have another string pointer in there:
1541 register char *pat = SvPVx(*++MARK, fromlen);
1542 register char *patend = pat + fromlen;
1548 And just before we start the loop, we'll set C<patcopy> to be the start
1553 sv_setpvn(cat, "", 0);
1555 while (pat < patend) {
1557 Now if we see a C<U> which was at the start of the string, we turn on
1558 the C<UTF8> flag for the output SV, C<cat>:
1560 + if (datumtype == 'U' && pat==patcopy+1)
1562 if (datumtype == '#') {
1563 while (pat < patend && *pat != '\n')
1566 Remember that it has to be C<patcopy+1> because the first character of
1567 the string is the C<U> which has been swallowed into C<datumtype!>
1569 Oops, we forgot one thing: what if there are spaces at the start of the
1570 pattern? C<pack(" U*", @stuff)> will have C<U> as the first active
1571 character, even though it's not the first thing in the pattern. In this
1572 case, we have to advance C<patcopy> along with C<pat> when we see spaces:
1574 if (isSPACE(datumtype))
1579 if (isSPACE(datumtype)) {
1584 OK. That's the C part done. Now we must do two additional things before
1585 this patch is ready to go: we've changed the behaviour of Perl, and so
1586 we must document that change. We must also provide some more regression
1587 tests to make sure our patch works and doesn't create a bug somewhere
1588 else along the line.
1590 The regression tests for each operator live in F<t/op/>, and so we
1591 make a copy of F<t/op/pack.t> to F<t/op/pack.t~>. Now we can add our
1592 tests to the end. First, we'll test that the C<U> does indeed create
1595 t/op/pack.t has a sensible ok() function, but if it didn't we could
1596 use the one from t/test.pl.
1598 require './test.pl';
1599 plan( tests => 159 );
1603 print 'not ' unless "1.20.300.4000" eq sprintf "%vd", pack("U*",1,20,300,4000);
1604 print "ok $test\n"; $test++;
1606 we can write the more sensible (see L<Test::More> for a full
1607 explanation of is() and other testing functions).
1609 is( "1.20.300.4000", sprintf "%vd", pack("U*",1,20,300,4000),
1610 "U* produces Unicode" );
1612 Now we'll test that we got that space-at-the-beginning business right:
1614 is( "1.20.300.4000", sprintf "%vd", pack(" U*",1,20,300,4000),
1615 " with spaces at the beginning" );
1617 And finally we'll test that we don't make Unicode strings if C<U> is B<not>
1618 the first active format:
1620 isnt( v1.20.300.4000, sprintf "%vd", pack("C0U*",1,20,300,4000),
1621 "U* not first isn't Unicode" );
1623 Mustn't forget to change the number of tests which appears at the top,
1624 or else the automated tester will get confused. This will either look
1631 plan( tests => 156 );
1633 We now compile up Perl, and run it through the test suite. Our new
1636 Finally, the documentation. The job is never done until the paperwork is
1637 over, so let's describe the change we've just made. The relevant place
1638 is F<pod/perlfunc.pod>; again, we make a copy, and then we'll insert
1639 this text in the description of C<pack>:
1643 If the pattern begins with a C<U>, the resulting string will be treated
1644 as UTF-8-encoded Unicode. You can force UTF-8 encoding on in a string
1645 with an initial C<U0>, and the bytes that follow will be interpreted as
1646 Unicode characters. If you don't want this to happen, you can begin your
1647 pattern with C<C0> (or anything else) to force Perl not to UTF-8 encode your
1648 string, and then follow this with a C<U*> somewhere in your pattern.
1650 All done. Now let's create the patch. F<Porting/patching.pod> tells us
1651 that if we're making major changes, we should copy the entire directory
1652 to somewhere safe before we begin fiddling, and then do
1654 diff -ruN old new > patch
1656 However, we know which files we've changed, and we can simply do this:
1658 diff -u pp.c~ pp.c > patch
1659 diff -u t/op/pack.t~ t/op/pack.t >> patch
1660 diff -u pod/perlfunc.pod~ pod/perlfunc.pod >> patch
1662 We end up with a patch looking a little like this:
1664 --- pp.c~ Fri Jun 02 04:34:10 2000
1665 +++ pp.c Fri Jun 16 11:37:25 2000
1666 @@ -4375,6 +4375,7 @@
1669 register char *pat = SvPVx(*++MARK, fromlen);
1671 register char *patend = pat + fromlen;
1674 @@ -4405,6 +4406,7 @@
1677 And finally, we submit it, with our rationale, to perl5-porters. Job
1680 =head2 Patching a core module
1682 This works just like patching anything else, with an extra
1683 consideration. Many core modules also live on CPAN. If this is so,
1684 patch the CPAN version instead of the core and send the patch off to
1685 the module maintainer (with a copy to p5p). This will help the module
1686 maintainer keep the CPAN version in sync with the core version without
1687 constantly scanning p5p.
1689 The list of maintainers of core modules is usefully documented in
1690 F<Porting/Maintainers.pl>.
1692 =head2 Adding a new function to the core
1694 If, as part of a patch to fix a bug, or just because you have an
1695 especially good idea, you decide to add a new function to the core,
1696 discuss your ideas on p5p well before you start work. It may be that
1697 someone else has already attempted to do what you are considering and
1698 can give lots of good advice or even provide you with bits of code
1699 that they already started (but never finished).
1701 You have to follow all of the advice given above for patching. It is
1702 extremely important to test any addition thoroughly and add new tests
1703 to explore all boundary conditions that your new function is expected
1704 to handle. If your new function is used only by one module (e.g. toke),
1705 then it should probably be named S_your_function (for static); on the
1706 other hand, if you expect it to accessible from other functions in
1707 Perl, you should name it Perl_your_function. See L<perlguts/Internal Functions>
1710 The location of any new code is also an important consideration. Don't
1711 just create a new top level .c file and put your code there; you would
1712 have to make changes to Configure (so the Makefile is created properly),
1713 as well as possibly lots of include files. This is strictly pumpking
1716 It is better to add your function to one of the existing top level
1717 source code files, but your choice is complicated by the nature of
1718 the Perl distribution. Only the files that are marked as compiled
1719 static are located in the perl executable. Everything else is located
1720 in the shared library (or DLL if you are running under WIN32). So,
1721 for example, if a function was only used by functions located in
1722 toke.c, then your code can go in toke.c. If, however, you want to call
1723 the function from universal.c, then you should put your code in another
1724 location, for example util.c.
1726 In addition to writing your c-code, you will need to create an
1727 appropriate entry in embed.pl describing your function, then run
1728 'make regen_headers' to create the entries in the numerous header
1729 files that perl needs to compile correctly. See L<perlguts/Internal Functions>
1730 for information on the various options that you can set in embed.pl.
1731 You will forget to do this a few (or many) times and you will get
1732 warnings during the compilation phase. Make sure that you mention
1733 this when you post your patch to P5P; the pumpking needs to know this.
1735 When you write your new code, please be conscious of existing code
1736 conventions used in the perl source files. See L<perlstyle> for
1737 details. Although most of the guidelines discussed seem to focus on
1738 Perl code, rather than c, they all apply (except when they don't ;).
1739 See also I<Porting/patching.pod> file in the Perl source distribution
1740 for lots of details about both formatting and submitting patches of
1743 Lastly, TEST TEST TEST TEST TEST any code before posting to p5p.
1744 Test on as many platforms as you can find. Test as many perl
1745 Configure options as you can (e.g. MULTIPLICITY). If you have
1746 profiling or memory tools, see L<EXTERNAL TOOLS FOR DEBUGGING PERL>
1747 below for how to use them to further test your code. Remember that
1748 most of the people on P5P are doing this on their own time and
1749 don't have the time to debug your code.
1751 =head2 Writing a test
1753 Every module and built-in function has an associated test file (or
1754 should...). If you add or change functionality, you have to write a
1755 test. If you fix a bug, you have to write a test so that bug never
1756 comes back. If you alter the docs, it would be nice to test what the
1757 new documentation says.
1759 In short, if you submit a patch you probably also have to patch the
1762 For modules, the test file is right next to the module itself.
1763 F<lib/strict.t> tests F<lib/strict.pm>. This is a recent innovation,
1764 so there are some snags (and it would be wonderful for you to brush
1765 them out), but it basically works that way. Everything else lives in
1768 If you add a new test directory under F<t/>, it is imperative that you
1769 add that directory to F<t/HARNESS> and F<t/TEST>.
1775 Testing of the absolute basic functionality of Perl. Things like
1776 C<if>, basic file reads and writes, simple regexes, etc. These are
1777 run first in the test suite and if any of them fail, something is
1782 These test the basic control structures, C<if/else>, C<while>,
1787 Tests basic issues of how Perl parses and compiles itself.
1791 Tests for built-in IO functions, including command line arguments.
1795 The old home for the module tests, you shouldn't put anything new in
1796 here. There are still some bits and pieces hanging around in here
1797 that need to be moved. Perhaps you could move them? Thanks!
1801 Tests for perl's method resolution order implementations
1806 Tests for perl's built in functions that don't fit into any of the
1811 Tests for POD directives. There are still some tests for the Pod
1812 modules hanging around in here that need to be moved out into F<lib/>.
1816 Tests for regex related functions or behaviour. (These used to live
1821 Testing features of how perl actually runs, including exit codes and
1822 handling of PERL* environment variables.
1826 Tests for the core support of Unicode.
1830 Windows-specific tests.
1834 A test suite for the s2p converter.
1838 The core uses the same testing style as the rest of Perl, a simple
1839 "ok/not ok" run through Test::Harness, but there are a few special
1842 There are three ways to write a test in the core. Test::More,
1843 t/test.pl and ad hoc C<print $test ? "ok 42\n" : "not ok 42\n">. The
1844 decision of which to use depends on what part of the test suite you're
1845 working on. This is a measure to prevent a high-level failure (such
1846 as Config.pm breaking) from causing basic functionality tests to fail.
1852 Since we don't know if require works, or even subroutines, use ad hoc
1853 tests for these two. Step carefully to avoid using the feature being
1856 =item t/cmd t/run t/io t/op
1858 Now that basic require() and subroutines are tested, you can use the
1859 t/test.pl library which emulates the important features of Test::More
1860 while using a minimum of core features.
1862 You can also conditionally use certain libraries like Config, but be
1863 sure to skip the test gracefully if it's not there.
1867 Now that the core of Perl is tested, Test::More can be used. You can
1868 also use the full suite of core modules in the tests.
1872 When you say "make test" Perl uses the F<t/TEST> program to run the
1873 test suite (except under Win32 where it uses F<t/harness> instead.)
1874 All tests are run from the F<t/> directory, B<not> the directory
1875 which contains the test. This causes some problems with the tests
1876 in F<lib/>, so here's some opportunity for some patching.
1878 You must be triply conscious of cross-platform concerns. This usually
1879 boils down to using File::Spec and avoiding things like C<fork()> and
1880 C<system()> unless absolutely necessary.
1882 =head2 Special Make Test Targets
1884 There are various special make targets that can be used to test Perl
1885 slightly differently than the standard "test" target. Not all them
1886 are expected to give a 100% success rate. Many of them have several
1887 aliases, and many of them are not available on certain operating
1894 Run F<perl> on all core tests (F<t/*> and F<lib/[a-z]*> pragma tests).
1896 (Not available on Win32)
1900 Run all the tests through B::Deparse. Not all tests will succeed.
1902 (Not available on Win32)
1904 =item test.taintwarn
1906 Run all tests with the B<-t> command-line switch. Not all tests
1907 are expected to succeed (until they're specifically fixed, of course).
1909 (Not available on Win32)
1913 Run F<miniperl> on F<t/base>, F<t/comp>, F<t/cmd>, F<t/run>, F<t/io>,
1914 F<t/op>, F<t/uni> and F<t/mro> tests.
1916 =item test.valgrind check.valgrind utest.valgrind ucheck.valgrind
1918 (Only in Linux) Run all the tests using the memory leak + naughty
1919 memory access tool "valgrind". The log files will be named
1920 F<testname.valgrind>.
1922 =item test.third check.third utest.third ucheck.third
1924 (Only in Tru64) Run all the tests using the memory leak + naughty
1925 memory access tool "Third Degree". The log files will be named
1926 F<perl.3log.testname>.
1928 =item test.torture torturetest
1930 Run all the usual tests and some extra tests. As of Perl 5.8.0 the
1931 only extra tests are Abigail's JAPHs, F<t/japh/abigail.t>.
1933 You can also run the torture test with F<t/harness> by giving
1934 C<-torture> argument to F<t/harness>.
1936 =item utest ucheck test.utf8 check.utf8
1938 Run all the tests with -Mutf8. Not all tests will succeed.
1940 (Not available on Win32)
1942 =item minitest.utf16 test.utf16
1944 Runs the tests with UTF-16 encoded scripts, encoded with different
1945 versions of this encoding.
1947 C<make utest.utf16> runs the test suite with a combination of C<-utf8> and
1948 C<-utf16> arguments to F<t/TEST>.
1950 (Not available on Win32)
1954 Run the test suite with the F<t/harness> controlling program, instead of
1955 F<t/TEST>. F<t/harness> is more sophisticated, and uses the
1956 L<Test::Harness> module, thus using this test target supposes that perl
1957 mostly works. The main advantage for our purposes is that it prints a
1958 detailed summary of failed tests at the end. Also, unlike F<t/TEST>, it
1959 doesn't redirect stderr to stdout.
1961 Note that under Win32 F<t/harness> is always used instead of F<t/TEST>, so
1962 there is no special "test_harness" target.
1964 Under Win32's "test" target you may use the TEST_SWITCHES and TEST_FILES
1965 environment variables to control the behaviour of F<t/harness>. This means
1968 nmake test TEST_FILES="op/*.t"
1969 nmake test TEST_SWITCHES="-torture" TEST_FILES="op/*.t"
1971 =item Parallel tests
1973 The core distribution can now run its regression tests in parallel on
1974 Unix-like platforms. Instead of running C<make test>, set C<TEST_JOBS> in
1975 your environment to the number of tests to run in parallel, and run
1976 C<make test_harness>. On a Bourne-like shell, this can be done as
1978 TEST_JOBS=3 make test_harness # Run 3 tests in parallel
1980 An environment variable is used, rather than parallel make itself, because
1981 L<TAP::Harness> needs to be able to schedule individual non-conflicting test
1982 scripts itself, and there is no standard interface to C<make> utilities to
1983 interact with their job schedulers.
1985 Note that currently some test scripts may fail when run in parallel (most
1986 notably C<ext/IO/t/io_dir.t>). If necessary run just the failing scripts
1987 again sequentially and see if the failures go away.
1988 =item test-notty test_notty
1990 Sets PERL_SKIP_TTY_TEST to true before running normal test.
1994 =head2 Running tests by hand
1996 You can run part of the test suite by hand by using one the following
1997 commands from the F<t/> directory :
1999 ./perl -I../lib TEST list-of-.t-files
2003 ./perl -I../lib harness list-of-.t-files
2005 (if you don't specify test scripts, the whole test suite will be run.)
2007 =head3 Using t/harness for testing
2009 If you use C<harness> for testing you have several command line options
2010 available to you. The arguments are as follows, and are in the order
2011 that they must appear if used together.
2013 harness -v -torture -re=pattern LIST OF FILES TO TEST
2014 harness -v -torture -re LIST OF PATTERNS TO MATCH
2016 If C<LIST OF FILES TO TEST> is omitted the file list is obtained from
2017 the manifest. The file list may include shell wildcards which will be
2024 Run the tests under verbose mode so you can see what tests were run,
2029 Run the torture tests as well as the normal set.
2033 Filter the file list so that all the test files run match PATTERN.
2034 Note that this form is distinct from the B<-re LIST OF PATTERNS> form below
2035 in that it allows the file list to be provided as well.
2037 =item -re LIST OF PATTERNS
2039 Filter the file list so that all the test files run match
2040 /(LIST|OF|PATTERNS)/. Note that with this form the patterns
2041 are joined by '|' and you cannot supply a list of files, instead
2042 the test files are obtained from the MANIFEST.
2046 You can run an individual test by a command similar to
2048 ./perl -I../lib patho/to/foo.t
2050 except that the harnesses set up some environment variables that may
2051 affect the execution of the test :
2057 indicates that we're running this test part of the perl core test suite.
2058 This is useful for modules that have a dual life on CPAN.
2060 =item PERL_DESTRUCT_LEVEL=2
2062 is set to 2 if it isn't set already (see L</PERL_DESTRUCT_LEVEL>)
2066 (used only by F<t/TEST>) if set, overrides the path to the perl executable
2067 that should be used to run the tests (the default being F<./perl>).
2069 =item PERL_SKIP_TTY_TEST
2071 if set, tells to skip the tests that need a terminal. It's actually set
2072 automatically by the Makefile, but can also be forced artificially by
2073 running 'make test_notty'.
2077 =head3 Other environment variables that may influence tests
2081 =item PERL_TEST_Net_Ping
2083 Setting this variable runs all the Net::Ping modules tests,
2084 otherwise some tests that interact with the outside world are skipped.
2087 =item PERL_TEST_NOVREXX
2089 Setting this variable skips the vrexx.t tests for OS2::REXX.
2091 =item PERL_TEST_NUMCONVERTS
2093 This sets a variable in op/numconvert.t.
2097 See also the documentation for the Test and Test::Harness modules,
2098 for more environment variables that affect testing.
2100 =head2 Common problems when patching Perl source code
2102 Perl source plays by ANSI C89 rules: no C99 (or C++) extensions. In
2103 some cases we have to take pre-ANSI requirements into consideration.
2104 You don't care about some particular platform having broken Perl?
2105 I hear there is still a strong demand for J2EE programmers.
2107 =head2 Perl environment problems
2113 Not compiling with threading
2115 Compiling with threading (-Duseithreads) completely rewrites
2116 the function prototypes of Perl. You better try your changes
2117 with that. Related to this is the difference between "Perl_-less"
2118 and "Perl_-ly" APIs, for example:
2120 Perl_sv_setiv(aTHX_ ...);
2123 The first one explicitly passes in the context, which is needed for e.g.
2124 threaded builds. The second one does that implicitly; do not get them
2125 mixed. If you are not passing in a aTHX_, you will need to do a dTHX
2126 (or a dVAR) as the first thing in the function.
2128 See L<perlguts/"How multiple interpreters and concurrency are supported">
2129 for further discussion about context.
2133 Not compiling with -DDEBUGGING
2135 The DEBUGGING define exposes more code to the compiler,
2136 therefore more ways for things to go wrong. You should try it.
2140 Introducing (non-read-only) globals
2142 Do not introduce any modifiable globals, truly global or file static.
2143 They are bad form and complicate multithreading and other forms of
2144 concurrency. The right way is to introduce them as new interpreter
2145 variables, see F<intrpvar.h> (at the very end for binary compatibility).
2147 Introducing read-only (const) globals is okay, as long as you verify
2148 with e.g. C<nm libperl.a|egrep -v ' [TURtr] '> (if your C<nm> has
2149 BSD-style output) that the data you added really is read-only.
2150 (If it is, it shouldn't show up in the output of that command.)
2152 If you want to have static strings, make them constant:
2154 static const char etc[] = "...";
2156 If you want to have arrays of constant strings, note carefully
2157 the right combination of C<const>s:
2159 static const char * const yippee[] =
2160 {"hi", "ho", "silver"};
2162 There is a way to completely hide any modifiable globals (they are all
2163 moved to heap), the compilation setting C<-DPERL_GLOBAL_STRUCT_PRIVATE>.
2164 It is not normally used, but can be used for testing, read more
2165 about it in L<perlguts/"Background and PERL_IMPLICIT_CONTEXT">.
2169 Not exporting your new function
2171 Some platforms (Win32, AIX, VMS, OS/2, to name a few) require any
2172 function that is part of the public API (the shared Perl library)
2173 to be explicitly marked as exported. See the discussion about
2174 F<embed.pl> in L<perlguts>.
2178 Exporting your new function
2180 The new shiny result of either genuine new functionality or your
2181 arduous refactoring is now ready and correctly exported. So what
2182 could possibly go wrong?
2184 Maybe simply that your function did not need to be exported in the
2185 first place. Perl has a long and not so glorious history of exporting
2186 functions that it should not have.
2188 If the function is used only inside one source code file, make it
2189 static. See the discussion about F<embed.pl> in L<perlguts>.
2191 If the function is used across several files, but intended only for
2192 Perl's internal use (and this should be the common case), do not
2193 export it to the public API. See the discussion about F<embed.pl>
2198 =head2 Portability problems
2200 The following are common causes of compilation and/or execution
2201 failures, not common to Perl as such. The C FAQ is good bedtime
2202 reading. Please test your changes with as many C compilers and
2203 platforms as possible -- we will, anyway, and it's nice to save
2204 oneself from public embarrassment.
2206 If using gcc, you can add the C<-std=c89> option which will hopefully
2207 catch most of these unportabilities. (However it might also catch
2208 incompatibilities in your system's header files.)
2210 Use the Configure C<-Dgccansipedantic> flag to enable the gcc
2211 C<-ansi -pedantic> flags which enforce stricter ANSI rules.
2213 If using the C<gcc -Wall> note that not all the possible warnings
2214 (like C<-Wunitialized>) are given unless you also compile with C<-O>.
2216 Note that if using gcc, starting from Perl 5.9.5 the Perl core source
2217 code files (the ones at the top level of the source code distribution,
2218 but not e.g. the extensions under ext/) are automatically compiled
2219 with as many as possible of the C<-std=c89>, C<-ansi>, C<-pedantic>,
2220 and a selection of C<-W> flags (see cflags.SH).
2222 Also study L<perlport> carefully to avoid any bad assumptions
2223 about the operating system, filesystems, and so forth.
2225 You may once in a while try a "make microperl" to see whether we
2226 can still compile Perl with just the bare minimum of interfaces.
2229 Do not assume an operating system indicates a certain compiler.
2235 Casting pointers to integers or casting integers to pointers
2237 void castaway(U8* p)
2243 void castaway(U8* p)
2247 Both are bad, and broken, and unportable. Use the PTR2IV()
2248 macro that does it right. (Likewise, there are PTR2UV(), PTR2NV(),
2249 INT2PTR(), and NUM2PTR().)
2253 Casting between data function pointers and data pointers
2255 Technically speaking casting between function pointers and data
2256 pointers is unportable and undefined, but practically speaking
2257 it seems to work, but you should use the FPTR2DPTR() and DPTR2FPTR()
2258 macros. Sometimes you can also play games with unions.
2262 Assuming sizeof(int) == sizeof(long)
2264 There are platforms where longs are 64 bits, and platforms where ints
2265 are 64 bits, and while we are out to shock you, even platforms where
2266 shorts are 64 bits. This is all legal according to the C standard.
2267 (In other words, "long long" is not a portable way to specify 64 bits,
2268 and "long long" is not even guaranteed to be any wider than "long".)
2270 Instead, use the definitions IV, UV, IVSIZE, I32SIZE, and so forth.
2271 Avoid things like I32 because they are B<not> guaranteed to be
2272 I<exactly> 32 bits, they are I<at least> 32 bits, nor are they
2273 guaranteed to be B<int> or B<long>. If you really explicitly need
2274 64-bit variables, use I64 and U64, but only if guarded by HAS_QUAD.
2278 Assuming one can dereference any type of pointer for any type of data
2281 long pony = *p; /* BAD */
2283 Many platforms, quite rightly so, will give you a core dump instead
2284 of a pony if the p happens not be correctly aligned.
2290 (int)*p = ...; /* BAD */
2292 Simply not portable. Get your lvalue to be of the right type,
2293 or maybe use temporary variables, or dirty tricks with unions.
2297 Assume B<anything> about structs (especially the ones you
2298 don't control, like the ones coming from the system headers)
2304 That a certain field exists in a struct
2308 That no other fields exist besides the ones you know of
2312 That a field is of certain signedness, sizeof, or type
2316 That the fields are in a certain order
2322 While C guarantees the ordering specified in the struct definition,
2323 between different platforms the definitions might differ
2329 That the sizeof(struct) or the alignments are the same everywhere
2335 There might be padding bytes between the fields to align the fields -
2336 the bytes can be anything
2340 Structs are required to be aligned to the maximum alignment required
2341 by the fields - which for native types is for usually equivalent to
2342 sizeof() of the field
2350 Assuming the character set is ASCIIish
2352 Perl can compile and run under EBCDIC platforms. See L<perlebcdic>.
2353 This is transparent for the most part, but because the character sets
2354 differ, you shouldn't use numeric (decimal, octal, nor hex) constants
2355 to refer to characters. You can safely say 'A', but not 0x41.
2356 You can safely say '\n', but not \012.
2357 If a character doesn't have a trivial input form, you can
2358 create a #define for it in both C<utfebcdic.h> and C<utf8.h>, so that
2359 it resolves to different values depending on the character set being used.
2360 (There are three different EBCDIC character sets defined in C<utfebcdic.h>,
2361 so it might be best to insert the #define three times in that file.)
2363 Also, the range 'A' - 'Z' in ASCII is an unbroken sequence of 26 upper case
2364 alphabetic characters. That is not true in EBCDIC. Nor for 'a' to 'z'.
2365 But '0' - '9' is an unbroken range in both systems. Don't assume anything
2368 Many of the comments in the existing code ignore the possibility of EBCDIC,
2369 and may be wrong therefore, even if the code works.
2370 This is actually a tribute to the successful transparent insertion of being
2371 able to handle EBCDIC without having to change pre-existing code.
2373 UTF-8 and UTF-EBCDIC are two different encodings used to represent Unicode
2374 code points as sequences of bytes. Macros
2375 with the same names (but different definitions)
2376 in C<utf8.h> and C<utfebcdic.h>
2377 are used to allow the calling code to think that there is only one such
2379 This is almost always referred to as C<utf8>, but it means the EBCDIC version
2380 as well. Again, comments in the code may well be wrong even if the code itself
2382 For example, the concept of C<invariant characters> differs between ASCII and
2384 On ASCII platforms, only characters that do not have the high-order
2385 bit set (i.e. whose ordinals are strict ASCII, 0 - 127)
2386 are invariant, and the documentation and comments in the code
2388 often referring to something like, say, C<hibit>.
2389 The situation differs and is not so simple on EBCDIC machines, but as long as
2390 the code itself uses the C<NATIVE_IS_INVARIANT()> macro appropriately, it
2391 works, even if the comments are wrong.
2395 Assuming the character set is just ASCII
2397 ASCII is a 7 bit encoding, but bytes have 8 bits in them. The 128 extra
2398 characters have different meanings depending on the locale. Absent a locale,
2399 currently these extra characters are generally considered to be unassigned,
2400 and this has presented some problems.
2401 This is scheduled to be changed in 5.12 so that these characters will
2402 be considered to be Latin-1 (ISO-8859-1).
2406 Mixing #define and #ifdef
2408 #define BURGLE(x) ... \
2409 #ifdef BURGLE_OLD_STYLE /* BAD */
2410 ... do it the old way ... \
2412 ... do it the new way ... \
2415 You cannot portably "stack" cpp directives. For example in the above
2416 you need two separate BURGLE() #defines, one for each #ifdef branch.
2420 Adding non-comment stuff after #endif or #else
2424 #else !SNOSH /* BAD */
2426 #endif SNOSH /* BAD */
2428 The #endif and #else cannot portably have anything non-comment after
2429 them. If you want to document what is going (which is a good idea
2430 especially if the branches are long), use (C) comments:
2438 The gcc option C<-Wendif-labels> warns about the bad variant
2439 (by default on starting from Perl 5.9.4).
2443 Having a comma after the last element of an enum list
2451 is not portable. Leave out the last comma.
2453 Also note that whether enums are implicitly morphable to ints
2454 varies between compilers, you might need to (int).
2460 // This function bamfoodles the zorklator. /* BAD */
2462 That is C99 or C++. Perl is C89. Using the //-comments is silently
2463 allowed by many C compilers but cranking up the ANSI C89 strictness
2464 (which we like to do) causes the compilation to fail.
2468 Mixing declarations and code
2473 set_zorkmids(n); /* BAD */
2476 That is C99 or C++. Some C compilers allow that, but you shouldn't.
2478 The gcc option C<-Wdeclaration-after-statements> scans for such problems
2479 (by default on starting from Perl 5.9.4).
2483 Introducing variables inside for()
2485 for(int i = ...; ...; ...) { /* BAD */
2487 That is C99 or C++. While it would indeed be awfully nice to have that
2488 also in C89, to limit the scope of the loop variable, alas, we cannot.
2492 Mixing signed char pointers with unsigned char pointers
2494 int foo(char *s) { ... }
2496 unsigned char *t = ...; /* Or U8* t = ... */
2499 While this is legal practice, it is certainly dubious, and downright
2500 fatal in at least one platform: for example VMS cc considers this a
2501 fatal error. One cause for people often making this mistake is that a
2502 "naked char" and therefore dereferencing a "naked char pointer" have
2503 an undefined signedness: it depends on the compiler and the flags of
2504 the compiler and the underlying platform whether the result is signed
2505 or unsigned. For this very same reason using a 'char' as an array
2510 Macros that have string constants and their arguments as substrings of
2511 the string constants
2513 #define FOO(n) printf("number = %d\n", n) /* BAD */
2516 Pre-ANSI semantics for that was equivalent to
2518 printf("10umber = %d\10");
2520 which is probably not what you were expecting. Unfortunately at least
2521 one reasonably common and modern C compiler does "real backward
2522 compatibility" here, in AIX that is what still happens even though the
2523 rest of the AIX compiler is very happily C89.
2527 Using printf formats for non-basic C types
2530 printf("i = %d\n", i); /* BAD */
2532 While this might by accident work in some platform (where IV happens
2533 to be an C<int>), in general it cannot. IV might be something larger.
2534 Even worse the situation is with more specific types (defined by Perl's
2535 configuration step in F<config.h>):
2538 printf("who = %d\n", who); /* BAD */
2540 The problem here is that Uid_t might be not only not C<int>-wide
2541 but it might also be unsigned, in which case large uids would be
2542 printed as negative values.
2544 There is no simple solution to this because of printf()'s limited
2545 intelligence, but for many types the right format is available as
2546 with either 'f' or '_f' suffix, for example:
2548 IVdf /* IV in decimal */
2549 UVxf /* UV is hexadecimal */
2551 printf("i = %"IVdf"\n", i); /* The IVdf is a string constant. */
2553 Uid_t_f /* Uid_t in decimal */
2555 printf("who = %"Uid_t_f"\n", who);
2557 Or you can try casting to a "wide enough" type:
2559 printf("i = %"IVdf"\n", (IV)something_very_small_and_signed);
2561 Also remember that the C<%p> format really does require a void pointer:
2564 printf("p = %p\n", (void*)p);
2566 The gcc option C<-Wformat> scans for such problems.
2570 Blindly using variadic macros
2572 gcc has had them for a while with its own syntax, and C99 brought
2573 them with a standardized syntax. Don't use the former, and use
2574 the latter only if the HAS_C99_VARIADIC_MACROS is defined.
2578 Blindly passing va_list
2580 Not all platforms support passing va_list to further varargs (stdarg)
2581 functions. The right thing to do is to copy the va_list using the
2582 Perl_va_copy() if the NEED_VA_COPY is defined.
2586 Using gcc statement expressions
2588 val = ({...;...;...}); /* BAD */
2590 While a nice extension, it's not portable. The Perl code does
2591 admittedly use them if available to gain some extra speed
2592 (essentially as a funky form of inlining), but you shouldn't.
2596 Binding together several statements in a macro
2598 Use the macros STMT_START and STMT_END.
2606 Testing for operating systems or versions when should be testing for features
2608 #ifdef __FOONIX__ /* BAD */
2612 Unless you know with 100% certainty that quux() is only ever available
2613 for the "Foonix" operating system B<and> that is available B<and>
2614 correctly working for B<all> past, present, B<and> future versions of
2615 "Foonix", the above is very wrong. This is more correct (though still
2616 not perfect, because the below is a compile-time check):
2622 How does the HAS_QUUX become defined where it needs to be? Well, if
2623 Foonix happens to be UNIXy enough to be able to run the Configure
2624 script, and Configure has been taught about detecting and testing
2625 quux(), the HAS_QUUX will be correctly defined. In other platforms,
2626 the corresponding configuration step will hopefully do the same.
2628 In a pinch, if you cannot wait for Configure to be educated,
2629 or if you have a good hunch of where quux() might be available,
2630 you can temporarily try the following:
2632 #if (defined(__FOONIX__) || defined(__BARNIX__))
2642 But in any case, try to keep the features and operating systems separate.
2646 =head2 Problematic System Interfaces
2652 malloc(0), realloc(0), calloc(0, 0) are non-portable. To be portable
2653 allocate at least one byte. (In general you should rarely need to
2654 work at this low level, but instead use the various malloc wrappers.)
2658 snprintf() - the return type is unportable. Use my_snprintf() instead.
2662 =head2 Security problems
2664 Last but not least, here are various tips for safer coding.
2672 Or we will publicly ridicule you. Seriously.
2676 Do not use strcpy() or strcat() or strncpy() or strncat()
2678 Use my_strlcpy() and my_strlcat() instead: they either use the native
2679 implementation, or Perl's own implementation (borrowed from the public
2680 domain implementation of INN).
2684 Do not use sprintf() or vsprintf()
2686 If you really want just plain byte strings, use my_snprintf()
2687 and my_vsnprintf() instead, which will try to use snprintf() and
2688 vsnprintf() if those safer APIs are available. If you want something
2689 fancier than a plain byte string, use SVs and Perl_sv_catpvf().
2693 =head1 EXTERNAL TOOLS FOR DEBUGGING PERL
2695 Sometimes it helps to use external tools while debugging and
2696 testing Perl. This section tries to guide you through using
2697 some common testing and debugging tools with Perl. This is
2698 meant as a guide to interfacing these tools with Perl, not
2699 as any kind of guide to the use of the tools themselves.
2701 B<NOTE 1>: Running under memory debuggers such as Purify, valgrind, or
2702 Third Degree greatly slows down the execution: seconds become minutes,
2703 minutes become hours. For example as of Perl 5.8.1, the
2704 ext/Encode/t/Unicode.t takes extraordinarily long to complete under
2705 e.g. Purify, Third Degree, and valgrind. Under valgrind it takes more
2706 than six hours, even on a snappy computer-- the said test must be
2707 doing something that is quite unfriendly for memory debuggers. If you
2708 don't feel like waiting, that you can simply kill away the perl
2711 B<NOTE 2>: To minimize the number of memory leak false alarms (see
2712 L</PERL_DESTRUCT_LEVEL> for more information), you have to have
2713 environment variable PERL_DESTRUCT_LEVEL set to 2. The F<TEST>
2714 and harness scripts do that automatically. But if you are running
2715 some of the tests manually-- for csh-like shells:
2717 setenv PERL_DESTRUCT_LEVEL 2
2719 and for Bourne-type shells:
2721 PERL_DESTRUCT_LEVEL=2
2722 export PERL_DESTRUCT_LEVEL
2724 or in UNIXy environments you can also use the C<env> command:
2726 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib ...
2728 B<NOTE 3>: There are known memory leaks when there are compile-time
2729 errors within eval or require, seeing C<S_doeval> in the call stack
2730 is a good sign of these. Fixing these leaks is non-trivial,
2731 unfortunately, but they must be fixed eventually.
2733 B<NOTE 4>: L<DynaLoader> will not clean up after itself completely
2734 unless Perl is built with the Configure option
2735 C<-Accflags=-DDL_UNLOAD_ALL_AT_EXIT>.
2737 =head2 Rational Software's Purify
2739 Purify is a commercial tool that is helpful in identifying
2740 memory overruns, wild pointers, memory leaks and other such
2741 badness. Perl must be compiled in a specific way for
2742 optimal testing with Purify. Purify is available under
2743 Windows NT, Solaris, HP-UX, SGI, and Siemens Unix.
2745 =head2 Purify on Unix
2747 On Unix, Purify creates a new Perl binary. To get the most
2748 benefit out of Purify, you should create the perl to Purify
2751 sh Configure -Accflags=-DPURIFY -Doptimize='-g' \
2752 -Uusemymalloc -Dusemultiplicity
2754 where these arguments mean:
2758 =item -Accflags=-DPURIFY
2760 Disables Perl's arena memory allocation functions, as well as
2761 forcing use of memory allocation functions derived from the
2764 =item -Doptimize='-g'
2766 Adds debugging information so that you see the exact source
2767 statements where the problem occurs. Without this flag, all
2768 you will see is the source filename of where the error occurred.
2772 Disable Perl's malloc so that Purify can more closely monitor
2773 allocations and leaks. Using Perl's malloc will make Purify
2774 report most leaks in the "potential" leaks category.
2776 =item -Dusemultiplicity
2778 Enabling the multiplicity option allows perl to clean up
2779 thoroughly when the interpreter shuts down, which reduces the
2780 number of bogus leak reports from Purify.
2784 Once you've compiled a perl suitable for Purify'ing, then you
2789 which creates a binary named 'pureperl' that has been Purify'ed.
2790 This binary is used in place of the standard 'perl' binary
2791 when you want to debug Perl memory problems.
2793 As an example, to show any memory leaks produced during the
2794 standard Perl testset you would create and run the Purify'ed
2799 ../pureperl -I../lib harness
2801 which would run Perl on test.pl and report any memory problems.
2803 Purify outputs messages in "Viewer" windows by default. If
2804 you don't have a windowing environment or if you simply
2805 want the Purify output to unobtrusively go to a log file
2806 instead of to the interactive window, use these following
2807 options to output to the log file "perl.log":
2809 setenv PURIFYOPTIONS "-chain-length=25 -windows=no \
2810 -log-file=perl.log -append-logfile=yes"
2812 If you plan to use the "Viewer" windows, then you only need this option:
2814 setenv PURIFYOPTIONS "-chain-length=25"
2816 In Bourne-type shells:
2819 export PURIFYOPTIONS
2821 or if you have the "env" utility:
2823 env PURIFYOPTIONS="..." ../pureperl ...
2827 Purify on Windows NT instruments the Perl binary 'perl.exe'
2828 on the fly. There are several options in the makefile you
2829 should change to get the most use out of Purify:
2835 You should add -DPURIFY to the DEFINES line so the DEFINES
2836 line looks something like:
2838 DEFINES = -DWIN32 -D_CONSOLE -DNO_STRICT $(CRYPT_FLAG) -DPURIFY=1
2840 to disable Perl's arena memory allocation functions, as
2841 well as to force use of memory allocation functions derived
2842 from the system malloc.
2844 =item USE_MULTI = define
2846 Enabling the multiplicity option allows perl to clean up
2847 thoroughly when the interpreter shuts down, which reduces the
2848 number of bogus leak reports from Purify.
2850 =item #PERL_MALLOC = define
2852 Disable Perl's malloc so that Purify can more closely monitor
2853 allocations and leaks. Using Perl's malloc will make Purify
2854 report most leaks in the "potential" leaks category.
2858 Adds debugging information so that you see the exact source
2859 statements where the problem occurs. Without this flag, all
2860 you will see is the source filename of where the error occurred.
2864 As an example, to show any memory leaks produced during the
2865 standard Perl testset you would create and run Purify as:
2870 purify ../perl -I../lib harness
2872 which would instrument Perl in memory, run Perl on test.pl,
2873 then finally report any memory problems.
2877 The excellent valgrind tool can be used to find out both memory leaks
2878 and illegal memory accesses. As of version 3.3.0, Valgrind only
2879 supports Linux on x86, x86-64 and PowerPC. The special "test.valgrind"
2880 target can be used to run the tests under valgrind. Found errors
2881 and memory leaks are logged in files named F<testfile.valgrind>.
2883 Valgrind also provides a cachegrind tool, invoked on perl as:
2885 VG_OPTS=--tool=cachegrind make test.valgrind
2887 As system libraries (most notably glibc) are also triggering errors,
2888 valgrind allows to suppress such errors using suppression files. The
2889 default suppression file that comes with valgrind already catches a lot
2890 of them. Some additional suppressions are defined in F<t/perl.supp>.
2892 To get valgrind and for more information see
2894 http://developer.kde.org/~sewardj/
2896 =head2 Compaq's/Digital's/HP's Third Degree
2898 Third Degree is a tool for memory leak detection and memory access checks.
2899 It is one of the many tools in the ATOM toolkit. The toolkit is only
2900 available on Tru64 (formerly known as Digital UNIX formerly known as
2903 When building Perl, you must first run Configure with -Doptimize=-g
2904 and -Uusemymalloc flags, after that you can use the make targets
2905 "perl.third" and "test.third". (What is required is that Perl must be
2906 compiled using the C<-g> flag, you may need to re-Configure.)
2908 The short story is that with "atom" you can instrument the Perl
2909 executable to create a new executable called F<perl.third>. When the
2910 instrumented executable is run, it creates a log of dubious memory
2911 traffic in file called F<perl.3log>. See the manual pages of atom and
2912 third for more information. The most extensive Third Degree
2913 documentation is available in the Compaq "Tru64 UNIX Programmer's
2914 Guide", chapter "Debugging Programs with Third Degree".
2916 The "test.third" leaves a lot of files named F<foo_bar.3log> in the t/
2917 subdirectory. There is a problem with these files: Third Degree is so
2918 effective that it finds problems also in the system libraries.
2919 Therefore you should used the Porting/thirdclean script to cleanup
2920 the F<*.3log> files.
2922 There are also leaks that for given certain definition of a leak,
2923 aren't. See L</PERL_DESTRUCT_LEVEL> for more information.
2925 =head2 PERL_DESTRUCT_LEVEL
2927 If you want to run any of the tests yourself manually using e.g.
2928 valgrind, or the pureperl or perl.third executables, please note that
2929 by default perl B<does not> explicitly cleanup all the memory it has
2930 allocated (such as global memory arenas) but instead lets the exit()
2931 of the whole program "take care" of such allocations, also known as
2932 "global destruction of objects".
2934 There is a way to tell perl to do complete cleanup: set the
2935 environment variable PERL_DESTRUCT_LEVEL to a non-zero value.
2936 The t/TEST wrapper does set this to 2, and this is what you
2937 need to do too, if you don't want to see the "global leaks":
2938 For example, for "third-degreed" Perl:
2940 env PERL_DESTRUCT_LEVEL=2 ./perl.third -Ilib t/foo/bar.t
2942 (Note: the mod_perl apache module uses also this environment variable
2943 for its own purposes and extended its semantics. Refer to the mod_perl
2944 documentation for more information. Also, spawned threads do the
2945 equivalent of setting this variable to the value 1.)
2947 If, at the end of a run you get the message I<N scalars leaked>, you can
2948 recompile with C<-DDEBUG_LEAKING_SCALARS>, which will cause the addresses
2949 of all those leaked SVs to be dumped along with details as to where each
2950 SV was originally allocated. This information is also displayed by
2951 Devel::Peek. Note that the extra details recorded with each SV increases
2952 memory usage, so it shouldn't be used in production environments. It also
2953 converts C<new_SV()> from a macro into a real function, so you can use
2954 your favourite debugger to discover where those pesky SVs were allocated.
2956 If you see that you're leaking memory at runtime, but neither valgrind
2957 nor C<-DDEBUG_LEAKING_SCALARS> will find anything, you're probably
2958 leaking SVs that are still reachable and will be properly cleaned up
2959 during destruction of the interpreter. In such cases, using the C<-Dm>
2960 switch can point you to the source of the leak. If the executable was
2961 built with C<-DDEBUG_LEAKING_SCALARS>, C<-Dm> will output SV allocations
2962 in addition to memory allocations. Each SV allocation has a distinct
2963 serial number that will be written on creation and destruction of the SV.
2964 So if you're executing the leaking code in a loop, you need to look for
2965 SVs that are created, but never destroyed between each cycle. If such an
2966 SV is found, set a conditional breakpoint within C<new_SV()> and make it
2967 break only when C<PL_sv_serial> is equal to the serial number of the
2968 leaking SV. Then you will catch the interpreter in exactly the state
2969 where the leaking SV is allocated, which is sufficient in many cases to
2970 find the source of the leak.
2972 As C<-Dm> is using the PerlIO layer for output, it will by itself
2973 allocate quite a bunch of SVs, which are hidden to avoid recursion.
2974 You can bypass the PerlIO layer if you use the SV logging provided
2975 by C<-DPERL_MEM_LOG> instead.
2979 If compiled with C<-DPERL_MEM_LOG>, both memory and SV allocations go
2980 through logging functions, which is handy for breakpoint setting.
2982 Unless C<-DPERL_MEM_LOG_NOIMPL> is also compiled, the logging
2983 functions read $ENV{PERL_MEM_LOG} to determine whether to log the
2984 event, and if so how:
2986 $ENV{PERL_MEM_LOG} =~ /m/ Log all memory ops
2987 $ENV{PERL_MEM_LOG} =~ /s/ Log all SV ops
2988 $ENV{PERL_MEM_LOG} =~ /t/ include timestamp in Log
2989 $ENV{PERL_MEM_LOG} =~ /^(\d+)/ write to FD given (default is 2)
2991 Memory logging is somewhat similar to C<-Dm> but is independent of
2992 C<-DDEBUGGING>, and at a higher level; all uses of Newx(), Renew(),
2993 and Safefree() are logged with the caller's source code file and line
2994 number (and C function name, if supported by the C compiler). In
2995 contrast, C<-Dm> is directly at the point of C<malloc()>. SV logging
2998 Since the logging doesn't use PerlIO, all SV allocations are logged
2999 and no extra SV allocations are introduced by enabling the logging.
3000 If compiled with C<-DDEBUG_LEAKING_SCALARS>, the serial number for
3001 each SV allocation is also logged.
3005 Depending on your platform there are various of profiling Perl.
3007 There are two commonly used techniques of profiling executables:
3008 I<statistical time-sampling> and I<basic-block counting>.
3010 The first method takes periodically samples of the CPU program
3011 counter, and since the program counter can be correlated with the code
3012 generated for functions, we get a statistical view of in which
3013 functions the program is spending its time. The caveats are that very
3014 small/fast functions have lower probability of showing up in the
3015 profile, and that periodically interrupting the program (this is
3016 usually done rather frequently, in the scale of milliseconds) imposes
3017 an additional overhead that may skew the results. The first problem
3018 can be alleviated by running the code for longer (in general this is a
3019 good idea for profiling), the second problem is usually kept in guard
3020 by the profiling tools themselves.
3022 The second method divides up the generated code into I<basic blocks>.
3023 Basic blocks are sections of code that are entered only in the
3024 beginning and exited only at the end. For example, a conditional jump
3025 starts a basic block. Basic block profiling usually works by
3026 I<instrumenting> the code by adding I<enter basic block #nnnn>
3027 book-keeping code to the generated code. During the execution of the
3028 code the basic block counters are then updated appropriately. The
3029 caveat is that the added extra code can skew the results: again, the
3030 profiling tools usually try to factor their own effects out of the
3033 =head2 Gprof Profiling
3035 gprof is a profiling tool available in many UNIX platforms,
3036 it uses F<statistical time-sampling>.
3038 You can build a profiled version of perl called "perl.gprof" by
3039 invoking the make target "perl.gprof" (What is required is that Perl
3040 must be compiled using the C<-pg> flag, you may need to re-Configure).
3041 Running the profiled version of Perl will create an output file called
3042 F<gmon.out> is created which contains the profiling data collected
3043 during the execution.
3045 The gprof tool can then display the collected data in various ways.
3046 Usually gprof understands the following options:
3052 Suppress statically defined functions from the profile.
3056 Suppress the verbose descriptions in the profile.
3060 Exclude the given routine and its descendants from the profile.
3064 Display only the given routine and its descendants in the profile.
3068 Generate a summary file called F<gmon.sum> which then may be given
3069 to subsequent gprof runs to accumulate data over several runs.
3073 Display routines that have zero usage.
3077 For more detailed explanation of the available commands and output
3078 formats, see your own local documentation of gprof.
3082 $ sh Configure -des -Dusedevel -Doptimize='-pg' && make perl.gprof
3083 $ ./perl.gprof someprog # creates gmon.out in current directory
3084 $ gprof ./perl.gprof > out
3087 =head2 GCC gcov Profiling
3089 Starting from GCC 3.0 I<basic block profiling> is officially available
3092 You can build a profiled version of perl called F<perl.gcov> by
3093 invoking the make target "perl.gcov" (what is required that Perl must
3094 be compiled using gcc with the flags C<-fprofile-arcs
3095 -ftest-coverage>, you may need to re-Configure).
3097 Running the profiled version of Perl will cause profile output to be
3098 generated. For each source file an accompanying ".da" file will be
3101 To display the results you use the "gcov" utility (which should
3102 be installed if you have gcc 3.0 or newer installed). F<gcov> is
3103 run on source code files, like this
3107 which will cause F<sv.c.gcov> to be created. The F<.gcov> files
3108 contain the source code annotated with relative frequencies of
3109 execution indicated by "#" markers.
3111 Useful options of F<gcov> include C<-b> which will summarise the
3112 basic block, branch, and function call coverage, and C<-c> which
3113 instead of relative frequencies will use the actual counts. For
3114 more information on the use of F<gcov> and basic block profiling
3115 with gcc, see the latest GNU CC manual, as of GCC 3.0 see
3117 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc.html
3119 and its section titled "8. gcov: a Test Coverage Program"
3121 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc_8.html#SEC132
3125 $ sh Configure -des -Doptimize='-g' -Accflags='-fprofile-arcs -ftest-coverage' \
3126 -Aldflags='-fprofile-arcs -ftest-coverage' && make perl.gcov
3127 $ rm -f regexec.c.gcov regexec.gcda
3130 $ view regexec.c.gcov
3132 =head2 Pixie Profiling
3134 Pixie is a profiling tool available on IRIX and Tru64 (aka Digital
3135 UNIX aka DEC OSF/1) platforms. Pixie does its profiling using
3136 I<basic-block counting>.
3138 You can build a profiled version of perl called F<perl.pixie> by
3139 invoking the make target "perl.pixie" (what is required is that Perl
3140 must be compiled using the C<-g> flag, you may need to re-Configure).
3142 In Tru64 a file called F<perl.Addrs> will also be silently created,
3143 this file contains the addresses of the basic blocks. Running the
3144 profiled version of Perl will create a new file called "perl.Counts"
3145 which contains the counts for the basic block for that particular
3148 To display the results you use the F<prof> utility. The exact
3149 incantation depends on your operating system, "prof perl.Counts" in
3150 IRIX, and "prof -pixie -all -L. perl" in Tru64.
3152 In IRIX the following prof options are available:
3158 Reports the most heavily used lines in descending order of use.
3159 Useful for finding the hotspot lines.
3163 Groups lines by procedure, with procedures sorted in descending order of use.
3164 Within a procedure, lines are listed in source order.
3165 Useful for finding the hotspots of procedures.
3169 In Tru64 the following options are available:
3175 Procedures sorted in descending order by the number of cycles executed
3176 in each procedure. Useful for finding the hotspot procedures.
3177 (This is the default option.)
3181 Lines sorted in descending order by the number of cycles executed in
3182 each line. Useful for finding the hotspot lines.
3184 =item -i[nvocations]
3186 The called procedures are sorted in descending order by number of calls
3187 made to the procedures. Useful for finding the most used procedures.
3191 Grouped by procedure, sorted by cycles executed per procedure.
3192 Useful for finding the hotspots of procedures.
3196 The compiler emitted code for these lines, but the code was unexecuted.
3200 Unexecuted procedures.
3204 For further information, see your system's manual pages for pixie and prof.
3206 =head2 Miscellaneous tricks
3212 Those debugging perl with the DDD frontend over gdb may find the
3215 You can extend the data conversion shortcuts menu, so for example you
3216 can display an SV's IV value with one click, without doing any typing.
3217 To do that simply edit ~/.ddd/init file and add after:
3219 ! Display shortcuts.
3220 Ddd*gdbDisplayShortcuts: \
3221 /t () // Convert to Bin\n\
3222 /d () // Convert to Dec\n\
3223 /x () // Convert to Hex\n\
3224 /o () // Convert to Oct(\n\
3226 the following two lines:
3228 ((XPV*) (())->sv_any )->xpv_pv // 2pvx\n\
3229 ((XPVIV*) (())->sv_any )->xiv_iv // 2ivx
3231 so now you can do ivx and pvx lookups or you can plug there the
3232 sv_peek "conversion":
3234 Perl_sv_peek(my_perl, (SV*)()) // sv_peek
3236 (The my_perl is for threaded builds.)
3237 Just remember that every line, but the last one, should end with \n\
3239 Alternatively edit the init file interactively via:
3240 3rd mouse button -> New Display -> Edit Menu
3242 Note: you can define up to 20 conversion shortcuts in the gdb
3247 If you see in a debugger a memory area mysteriously full of 0xABABABAB
3248 or 0xEFEFEFEF, you may be seeing the effect of the Poison() macros,
3253 Under ithreads the optree is read only. If you want to enforce this, to check
3254 for write accesses from buggy code, compile with C<-DPL_OP_SLAB_ALLOC> to
3255 enable the OP slab allocator and C<-DPERL_DEBUG_READONLY_OPS> to enable code
3256 that allocates op memory via C<mmap>, and sets it read-only at run time.
3257 Any write access to an op results in a C<SIGBUS> and abort.
3259 This code is intended for development only, and may not be portable even to
3260 all Unix variants. Also, it is an 80% solution, in that it isn't able to make
3261 all ops read only. Specifically it
3267 Only sets read-only on all slabs of ops at C<CHECK> time, hence ops allocated
3268 later via C<require> or C<eval> will be re-write
3272 Turns an entire slab of ops read-write if the refcount of any op in the slab
3273 needs to be decreased.
3277 Turns an entire slab of ops read-write if any op from the slab is freed.
3281 It's not possible to turn the slabs to read-only after an action requiring
3282 read-write access, as either can happen during op tree building time, so
3283 there may still be legitimate write access.
3285 However, as an 80% solution it is still effective, as currently it catches
3286 a write access during the generation of F<Config.pm>, which means that we
3287 can't yet build F<perl> with this enabled.
3294 We've had a brief look around the Perl source, how to maintain quality
3295 of the source code, an overview of the stages F<perl> goes through
3296 when it's running your code, how to use debuggers to poke at the Perl
3297 guts, and finally how to analyse the execution of Perl. We took a very
3298 simple problem and demonstrated how to solve it fully - with
3299 documentation, regression tests, and finally a patch for submission to
3300 p5p. Finally, we talked about how to use external tools to debug and
3303 I'd now suggest you read over those references again, and then, as soon
3304 as possible, get your hands dirty. The best way to learn is by doing,
3311 Subscribe to perl5-porters, follow the patches and try and understand
3312 them; don't be afraid to ask if there's a portion you're not clear on -
3313 who knows, you may unearth a bug in the patch...
3317 Keep up to date with the bleeding edge Perl distributions and get
3318 familiar with the changes. Try and get an idea of what areas people are
3319 working on and the changes they're making.
3323 Do read the README associated with your operating system, e.g. README.aix
3324 on the IBM AIX OS. Don't hesitate to supply patches to that README if
3325 you find anything missing or changed over a new OS release.
3329 Find an area of Perl that seems interesting to you, and see if you can
3330 work out how it works. Scan through the source, and step over it in the
3331 debugger. Play, poke, investigate, fiddle! You'll probably get to
3332 understand not just your chosen area but a much wider range of F<perl>'s
3333 activity as well, and probably sooner than you'd think.
3339 =item I<The Road goes ever on and on, down from the door where it began.>
3343 If you can do these things, you've started on the long road to Perl porting.
3344 Thanks for wanting to help make Perl better - and happy hacking!
3346 =head2 Metaphoric Quotations
3348 If you recognized the quote about the Road above, you're in luck.
3350 Most software projects begin each file with a literal description of each
3351 file's purpose. Perl instead begins each with a literary allusion to that
3354 Like chapters in many books, all top-level Perl source files (along with a
3355 few others here and there) begin with an epigramic inscription that alludes,
3356 indirectly and metaphorically, to the material you're about to read.
3358 Quotations are taken from writings of J.R.R Tolkien pertaining to his
3359 Legendarium, almost always from I<The Lord of the Rings>. Chapters and
3360 page numbers are given using the following editions:
3366 I<The Hobbit>, by J.R.R. Tolkien. The hardcover, 70th-anniversary
3367 edition of 2007 was used, published in the UK by Harper Collins Publishers
3368 and in the US by the Houghton Mifflin Company.
3372 I<The Lord of the Rings>, by J.R.R. Tolkien. The hardcover,
3373 50th-anniversary edition of 2004 was used, published in the UK by Harper
3374 Collins Publishers and in the US by the Houghton Mifflin Company.
3378 I<The Lays of Beleriand>, by J.R.R. Tolkien and published posthumously by his
3379 son and literary executor, C.J.R. Tolkien, being the 3rd of the 12 volumes
3380 in Christopher's mammoth I<History of Middle Earth>. Page numbers derive
3381 from the hardcover edition, first published in 1983 by George Allen &
3382 Unwin; no page numbers changed for the special 3-volume omnibus edition of
3383 2002 or the various trade-paper editions, all again now by Harper Collins
3384 or Houghton Mifflin.
3388 Other JRRT books fair game for quotes would thus include I<The Adventures of
3389 Tom Bombadil>, I<The Silmarillion>, I<Unfinished Tales>, and I<The Tale of
3390 the Children of Hurin>, all but the first posthumously assembled by CJRT.
3391 But I<The Lord of the Rings> itself is perfectly fine and probably best to
3392 quote from, provided you can find a suitable quote there.
3394 So if you were to supply a new, complete, top-level source file to add to
3395 Perl, you should conform to this peculiar practice by yourself selecting an
3396 appropriate quotation from Tolkien, retaining the original spelling and
3397 punctuation and using the same format the rest of the quotes are in.
3398 Indirect and oblique is just fine; remember, it's a metaphor, so being meta
3399 is, after all, what it's for.
3403 This document was written by Nathan Torkington, and is maintained by
3404 the perl5-porters mailing list.