This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Run Porting/podtidy on pod/perlhacktips.pod
[perl5.git] / pod / perlhacktips.pod
CommitLineData
0061d4fa 1
04c692a8
DR
2=encoding utf8
3
4=for comment
5Consistent formatting of this file is achieved with:
6 perl ./Porting/podtidy pod/perlhacktips.pod
7
8=head1 NAME
9
10perlhacktips - Tips for Perl core C code hacking
11
12=head1 DESCRIPTION
13
14This document will help you learn the best way to go about hacking on
15the Perl core C code. It covers common problems, debugging, profiling,
16and more.
17
18If you haven't read L<perlhack> and L<perlhacktut> yet, you might want
19to do that first.
20
21=head1 COMMON PROBLEMS
22
23Perl source plays by ANSI C89 rules: no C99 (or C++) extensions. In
24some cases we have to take pre-ANSI requirements into consideration.
25You don't care about some particular platform having broken Perl? I
26hear there is still a strong demand for J2EE programmers.
27
28=head2 Perl environment problems
29
30=over 4
31
32=item *
33
34Not compiling with threading
35
36Compiling with threading (-Duseithreads) completely rewrites the
37function prototypes of Perl. You better try your changes with that.
38Related to this is the difference between "Perl_-less" and "Perl_-ly"
39APIs, for example:
40
41 Perl_sv_setiv(aTHX_ ...);
42 sv_setiv(...);
43
44The first one explicitly passes in the context, which is needed for
45e.g. threaded builds. The second one does that implicitly; do not get
46them mixed. If you are not passing in a aTHX_, you will need to do a
47dTHX (or a dVAR) as the first thing in the function.
48
49See L<perlguts/"How multiple interpreters and concurrency are
50supported"> for further discussion about context.
51
52=item *
53
54Not compiling with -DDEBUGGING
55
56The DEBUGGING define exposes more code to the compiler, therefore more
57ways for things to go wrong. You should try it.
58
59=item *
60
61Introducing (non-read-only) globals
62
63Do not introduce any modifiable globals, truly global or file static.
64They are bad form and complicate multithreading and other forms of
65concurrency. The right way is to introduce them as new interpreter
66variables, see F<intrpvar.h> (at the very end for binary
67compatibility).
68
69Introducing read-only (const) globals is okay, as long as you verify
70with e.g. C<nm libperl.a|egrep -v ' [TURtr] '> (if your C<nm> has
71BSD-style output) that the data you added really is read-only. (If it
72is, it shouldn't show up in the output of that command.)
73
74If you want to have static strings, make them constant:
75
76 static const char etc[] = "...";
77
78If you want to have arrays of constant strings, note carefully the
79right combination of C<const>s:
80
81 static const char * const yippee[] =
82 {"hi", "ho", "silver"};
83
84There is a way to completely hide any modifiable globals (they are all
85moved to heap), the compilation setting
86C<-DPERL_GLOBAL_STRUCT_PRIVATE>. It is not normally used, but can be
87used for testing, read more about it in L<perlguts/"Background and
88PERL_IMPLICIT_CONTEXT">.
89
90=item *
91
92Not exporting your new function
93
94Some platforms (Win32, AIX, VMS, OS/2, to name a few) require any
95function that is part of the public API (the shared Perl library) to be
96explicitly marked as exported. See the discussion about F<embed.pl> in
97L<perlguts>.
98
99=item *
100
101Exporting your new function
102
103The new shiny result of either genuine new functionality or your
104arduous refactoring is now ready and correctly exported. So what could
105possibly go wrong?
106
107Maybe simply that your function did not need to be exported in the
108first place. Perl has a long and not so glorious history of exporting
109functions that it should not have.
110
111If the function is used only inside one source code file, make it
112static. See the discussion about F<embed.pl> in L<perlguts>.
113
114If the function is used across several files, but intended only for
115Perl's internal use (and this should be the common case), do not export
116it to the public API. See the discussion about F<embed.pl> in
117L<perlguts>.
118
119=back
120
121=head2 Portability problems
122
123The following are common causes of compilation and/or execution
124failures, not common to Perl as such. The C FAQ is good bedtime
125reading. Please test your changes with as many C compilers and
126platforms as possible; we will, anyway, and it's nice to save oneself
127from public embarrassment.
128
129If using gcc, you can add the C<-std=c89> option which will hopefully
130catch most of these unportabilities. (However it might also catch
131incompatibilities in your system's header files.)
132
133Use the Configure C<-Dgccansipedantic> flag to enable the gcc C<-ansi
134-pedantic> flags which enforce stricter ANSI rules.
135
136If using the C<gcc -Wall> note that not all the possible warnings (like
137C<-Wunitialized>) are given unless you also compile with C<-O>.
138
139Note that if using gcc, starting from Perl 5.9.5 the Perl core source
140code files (the ones at the top level of the source code distribution,
141but not e.g. the extensions under ext/) are automatically compiled with
142as many as possible of the C<-std=c89>, C<-ansi>, C<-pedantic>, and a
143selection of C<-W> flags (see cflags.SH).
144
145Also study L<perlport> carefully to avoid any bad assumptions about the
146operating system, filesystems, and so forth.
147
148You may once in a while try a "make microperl" to see whether we can
149still compile Perl with just the bare minimum of interfaces. (See
150README.micro.)
151
152Do not assume an operating system indicates a certain compiler.
153
154=over 4
155
156=item *
157
158Casting pointers to integers or casting integers to pointers
159
160 void castaway(U8* p)
161 {
162 IV i = p;
163
164or
165
166 void castaway(U8* p)
167 {
168 IV i = (IV)p;
169
170Both are bad, and broken, and unportable. Use the PTR2IV() macro that
171does it right. (Likewise, there are PTR2UV(), PTR2NV(), INT2PTR(), and
172NUM2PTR().)
173
174=item *
175
176Casting between data function pointers and data pointers
177
178Technically speaking casting between function pointers and data
179pointers is unportable and undefined, but practically speaking it seems
180to work, but you should use the FPTR2DPTR() and DPTR2FPTR() macros.
181Sometimes you can also play games with unions.
182
183=item *
184
185Assuming sizeof(int) == sizeof(long)
186
187There are platforms where longs are 64 bits, and platforms where ints
188are 64 bits, and while we are out to shock you, even platforms where
189shorts are 64 bits. This is all legal according to the C standard. (In
190other words, "long long" is not a portable way to specify 64 bits, and
191"long long" is not even guaranteed to be any wider than "long".)
192
193Instead, use the definitions IV, UV, IVSIZE, I32SIZE, and so forth.
194Avoid things like I32 because they are B<not> guaranteed to be
195I<exactly> 32 bits, they are I<at least> 32 bits, nor are they
196guaranteed to be B<int> or B<long>. If you really explicitly need
19764-bit variables, use I64 and U64, but only if guarded by HAS_QUAD.
198
199=item *
200
201Assuming one can dereference any type of pointer for any type of data
202
203 char *p = ...;
204 long pony = *p; /* BAD */
205
206Many platforms, quite rightly so, will give you a core dump instead of
768312ab 207a pony if the p happens not to be correctly aligned.
04c692a8
DR
208
209=item *
210
211Lvalue casts
212
213 (int)*p = ...; /* BAD */
214
215Simply not portable. Get your lvalue to be of the right type, or maybe
216use temporary variables, or dirty tricks with unions.
217
218=item *
219
220Assume B<anything> about structs (especially the ones you don't
221control, like the ones coming from the system headers)
222
223=over 8
224
225=item *
226
227That a certain field exists in a struct
228
229=item *
230
231That no other fields exist besides the ones you know of
232
233=item *
234
235That a field is of certain signedness, sizeof, or type
236
237=item *
238
239That the fields are in a certain order
240
241=over 8
242
243=item *
244
245While C guarantees the ordering specified in the struct definition,
246between different platforms the definitions might differ
247
248=back
249
250=item *
251
252That the sizeof(struct) or the alignments are the same everywhere
253
254=over 8
255
256=item *
257
258There might be padding bytes between the fields to align the fields -
259the bytes can be anything
260
261=item *
262
263Structs are required to be aligned to the maximum alignment required by
264the fields - which for native types is for usually equivalent to
265sizeof() of the field
266
267=back
268
269=back
270
271=item *
272
273Assuming the character set is ASCIIish
274
275Perl can compile and run under EBCDIC platforms. See L<perlebcdic>.
276This is transparent for the most part, but because the character sets
277differ, you shouldn't use numeric (decimal, octal, nor hex) constants
278to refer to characters. You can safely say 'A', but not 0x41. You can
279safely say '\n', but not \012. If a character doesn't have a trivial
eb6d698b
KW
280input form, you should add it to the list in
281F<regen/unicode_constants.pl>, and have Perl create #defines for you,
282based on the current platform.
04c692a8
DR
283
284Also, the range 'A' - 'Z' in ASCII is an unbroken sequence of 26 upper
285case alphabetic characters. That is not true in EBCDIC. Nor for 'a' to
286'z'. But '0' - '9' is an unbroken range in both systems. Don't assume
287anything about other ranges.
288
289Many of the comments in the existing code ignore the possibility of
290EBCDIC, and may be wrong therefore, even if the code works. This is
291actually a tribute to the successful transparent insertion of being
292able to handle EBCDIC without having to change pre-existing code.
293
294UTF-8 and UTF-EBCDIC are two different encodings used to represent
295Unicode code points as sequences of bytes. Macros with the same names
296(but different definitions) in C<utf8.h> and C<utfebcdic.h> are used to
297allow the calling code to think that there is only one such encoding.
298This is almost always referred to as C<utf8>, but it means the EBCDIC
299version as well. Again, comments in the code may well be wrong even if
300the code itself is right. For example, the concept of C<invariant
301characters> differs between ASCII and EBCDIC. On ASCII platforms, only
302characters that do not have the high-order bit set (i.e. whose ordinals
303are strict ASCII, 0 - 127) are invariant, and the documentation and
304comments in the code may assume that, often referring to something
305like, say, C<hibit>. The situation differs and is not so simple on
306EBCDIC machines, but as long as the code itself uses the
307C<NATIVE_IS_INVARIANT()> macro appropriately, it works, even if the
308comments are wrong.
309
310=item *
311
312Assuming the character set is just ASCII
313
314ASCII is a 7 bit encoding, but bytes have 8 bits in them. The 128 extra
315characters have different meanings depending on the locale. Absent a
316locale, currently these extra characters are generally considered to be
317unassigned, and this has presented some problems. This is being changed
318starting in 5.12 so that these characters will be considered to be
319Latin-1 (ISO-8859-1).
320
321=item *
322
323Mixing #define and #ifdef
324
325 #define BURGLE(x) ... \
326 #ifdef BURGLE_OLD_STYLE /* BAD */
327 ... do it the old way ... \
328 #else
329 ... do it the new way ... \
330 #endif
331
332You cannot portably "stack" cpp directives. For example in the above
333you need two separate BURGLE() #defines, one for each #ifdef branch.
334
335=item *
336
337Adding non-comment stuff after #endif or #else
338
339 #ifdef SNOSH
340 ...
341 #else !SNOSH /* BAD */
342 ...
343 #endif SNOSH /* BAD */
344
345The #endif and #else cannot portably have anything non-comment after
346them. If you want to document what is going (which is a good idea
347especially if the branches are long), use (C) comments:
348
349 #ifdef SNOSH
350 ...
351 #else /* !SNOSH */
352 ...
353 #endif /* SNOSH */
354
355The gcc option C<-Wendif-labels> warns about the bad variant (by
356default on starting from Perl 5.9.4).
357
358=item *
359
360Having a comma after the last element of an enum list
361
362 enum color {
363 CERULEAN,
364 CHARTREUSE,
365 CINNABAR, /* BAD */
366 };
367
368is not portable. Leave out the last comma.
369
370Also note that whether enums are implicitly morphable to ints varies
371between compilers, you might need to (int).
372
373=item *
374
375Using //-comments
376
377 // This function bamfoodles the zorklator. /* BAD */
378
379That is C99 or C++. Perl is C89. Using the //-comments is silently
380allowed by many C compilers but cranking up the ANSI C89 strictness
381(which we like to do) causes the compilation to fail.
382
383=item *
384
385Mixing declarations and code
386
387 void zorklator()
388 {
389 int n = 3;
390 set_zorkmids(n); /* BAD */
391 int q = 4;
392
393That is C99 or C++. Some C compilers allow that, but you shouldn't.
394
395The gcc option C<-Wdeclaration-after-statements> scans for such
396problems (by default on starting from Perl 5.9.4).
397
398=item *
399
400Introducing variables inside for()
401
402 for(int i = ...; ...; ...) { /* BAD */
403
404That is C99 or C++. While it would indeed be awfully nice to have that
405also in C89, to limit the scope of the loop variable, alas, we cannot.
406
407=item *
408
409Mixing signed char pointers with unsigned char pointers
410
411 int foo(char *s) { ... }
412 ...
413 unsigned char *t = ...; /* Or U8* t = ... */
414 foo(t); /* BAD */
415
416While this is legal practice, it is certainly dubious, and downright
417fatal in at least one platform: for example VMS cc considers this a
418fatal error. One cause for people often making this mistake is that a
419"naked char" and therefore dereferencing a "naked char pointer" have an
420undefined signedness: it depends on the compiler and the flags of the
421compiler and the underlying platform whether the result is signed or
422unsigned. For this very same reason using a 'char' as an array index is
423bad.
424
425=item *
426
427Macros that have string constants and their arguments as substrings of
428the string constants
429
430 #define FOO(n) printf("number = %d\n", n) /* BAD */
431 FOO(10);
432
433Pre-ANSI semantics for that was equivalent to
434
435 printf("10umber = %d\10");
436
437which is probably not what you were expecting. Unfortunately at least
438one reasonably common and modern C compiler does "real backward
439compatibility" here, in AIX that is what still happens even though the
440rest of the AIX compiler is very happily C89.
441
442=item *
443
444Using printf formats for non-basic C types
445
446 IV i = ...;
447 printf("i = %d\n", i); /* BAD */
448
449While this might by accident work in some platform (where IV happens to
450be an C<int>), in general it cannot. IV might be something larger. Even
451worse the situation is with more specific types (defined by Perl's
452configuration step in F<config.h>):
453
454 Uid_t who = ...;
455 printf("who = %d\n", who); /* BAD */
456
457The problem here is that Uid_t might be not only not C<int>-wide but it
458might also be unsigned, in which case large uids would be printed as
459negative values.
460
461There is no simple solution to this because of printf()'s limited
462intelligence, but for many types the right format is available as with
463either 'f' or '_f' suffix, for example:
464
465 IVdf /* IV in decimal */
466 UVxf /* UV is hexadecimal */
467
468 printf("i = %"IVdf"\n", i); /* The IVdf is a string constant. */
469
470 Uid_t_f /* Uid_t in decimal */
471
472 printf("who = %"Uid_t_f"\n", who);
473
474Or you can try casting to a "wide enough" type:
475
476 printf("i = %"IVdf"\n", (IV)something_very_small_and_signed);
477
478Also remember that the C<%p> format really does require a void pointer:
479
480 U8* p = ...;
481 printf("p = %p\n", (void*)p);
482
483The gcc option C<-Wformat> scans for such problems.
484
485=item *
486
487Blindly using variadic macros
488
489gcc has had them for a while with its own syntax, and C99 brought them
490with a standardized syntax. Don't use the former, and use the latter
491only if the HAS_C99_VARIADIC_MACROS is defined.
492
493=item *
494
495Blindly passing va_list
496
497Not all platforms support passing va_list to further varargs (stdarg)
498functions. The right thing to do is to copy the va_list using the
499Perl_va_copy() if the NEED_VA_COPY is defined.
500
501=item *
502
503Using gcc statement expressions
504
505 val = ({...;...;...}); /* BAD */
506
507While a nice extension, it's not portable. The Perl code does
508admittedly use them if available to gain some extra speed (essentially
509as a funky form of inlining), but you shouldn't.
510
511=item *
512
513Binding together several statements in a macro
514
515Use the macros STMT_START and STMT_END.
516
517 STMT_START {
518 ...
519 } STMT_END
520
521=item *
522
523Testing for operating systems or versions when should be testing for
524features
525
526 #ifdef __FOONIX__ /* BAD */
527 foo = quux();
528 #endif
529
530Unless you know with 100% certainty that quux() is only ever available
531for the "Foonix" operating system B<and> that is available B<and>
532correctly working for B<all> past, present, B<and> future versions of
533"Foonix", the above is very wrong. This is more correct (though still
534not perfect, because the below is a compile-time check):
535
536 #ifdef HAS_QUUX
537 foo = quux();
538 #endif
539
540How does the HAS_QUUX become defined where it needs to be? Well, if
541Foonix happens to be Unixy enough to be able to run the Configure
542script, and Configure has been taught about detecting and testing
543quux(), the HAS_QUUX will be correctly defined. In other platforms, the
544corresponding configuration step will hopefully do the same.
545
546In a pinch, if you cannot wait for Configure to be educated, or if you
547have a good hunch of where quux() might be available, you can
548temporarily try the following:
549
550 #if (defined(__FOONIX__) || defined(__BARNIX__))
551 # define HAS_QUUX
552 #endif
553
554 ...
555
556 #ifdef HAS_QUUX
557 foo = quux();
558 #endif
559
560But in any case, try to keep the features and operating systems
561separate.
562
563=back
564
565=head2 Problematic System Interfaces
566
567=over 4
568
569=item *
570
571malloc(0), realloc(0), calloc(0, 0) are non-portable. To be portable
572allocate at least one byte. (In general you should rarely need to work
573at this low level, but instead use the various malloc wrappers.)
574
575=item *
576
577snprintf() - the return type is unportable. Use my_snprintf() instead.
578
579=back
580
581=head2 Security problems
582
583Last but not least, here are various tips for safer coding.
584
585=over 4
586
587=item *
588
589Do not use gets()
590
591Or we will publicly ridicule you. Seriously.
592
593=item *
594
595Do not use strcpy() or strcat() or strncpy() or strncat()
596
597Use my_strlcpy() and my_strlcat() instead: they either use the native
598implementation, or Perl's own implementation (borrowed from the public
599domain implementation of INN).
600
601=item *
602
603Do not use sprintf() or vsprintf()
604
605If you really want just plain byte strings, use my_snprintf() and
606my_vsnprintf() instead, which will try to use snprintf() and
607vsnprintf() if those safer APIs are available. If you want something
608fancier than a plain byte string, use SVs and Perl_sv_catpvf().
609
610=back
611
612=head1 DEBUGGING
613
614You can compile a special debugging version of Perl, which allows you
615to use the C<-D> option of Perl to tell more about what Perl is doing.
616But sometimes there is no alternative than to dive in with a debugger,
617either to see the stack trace of a core dump (very useful in a bug
618report), or trying to figure out what went wrong before the core dump
619happened, or how did we end up having wrong or unexpected results.
620
621=head2 Poking at Perl
622
623To really poke around with Perl, you'll probably want to build Perl for
624debugging, like this:
625
626 ./Configure -d -D optimize=-g
627 make
628
629C<-g> is a flag to the C compiler to have it produce debugging
630information which will allow us to step through a running program, and
631to see in which C function we are at (without the debugging information
632we might see only the numerical addresses of the functions, which is
633not very helpful).
634
635F<Configure> will also turn on the C<DEBUGGING> compilation symbol
636which enables all the internal debugging code in Perl. There are a
637whole bunch of things you can debug with this: L<perlrun> lists them
638all, and the best way to find out about them is to play about with
639them. The most useful options are probably
640
641 l Context (loop) stack processing
642 t Trace execution
643 o Method and overloading resolution
644 c String/numeric conversions
645
646Some of the functionality of the debugging code can be achieved using
647XS modules.
648
649 -Dr => use re 'debug'
650 -Dx => use O 'Debug'
651
652=head2 Using a source-level debugger
653
654If the debugging output of C<-D> doesn't help you, it's time to step
655through perl's execution with a source-level debugger.
656
657=over 3
658
659=item *
660
661We'll use C<gdb> for our examples here; the principles will apply to
662any debugger (many vendors call their debugger C<dbx>), but check the
663manual of the one you're using.
664
665=back
666
667To fire up the debugger, type
668
669 gdb ./perl
670
671Or if you have a core dump:
672
673 gdb ./perl core
674
675You'll want to do that in your Perl source tree so the debugger can
676read the source code. You should see the copyright message, followed by
677the prompt.
678
679 (gdb)
680
681C<help> will get you into the documentation, but here are the most
682useful commands:
683
684=over 3
685
686=item * run [args]
687
688Run the program with the given arguments.
689
690=item * break function_name
691
692=item * break source.c:xxx
693
694Tells the debugger that we'll want to pause execution when we reach
695either the named function (but see L<perlguts/Internal Functions>!) or
696the given line in the named source file.
697
698=item * step
699
700Steps through the program a line at a time.
701
702=item * next
703
704Steps through the program a line at a time, without descending into
705functions.
706
707=item * continue
708
709Run until the next breakpoint.
710
711=item * finish
712
713Run until the end of the current function, then stop again.
714
715=item * 'enter'
716
717Just pressing Enter will do the most recent operation again - it's a
718blessing when stepping through miles of source code.
719
720=item * print
721
722Execute the given C code and print its results. B<WARNING>: Perl makes
723heavy use of macros, and F<gdb> does not necessarily support macros
724(see later L</"gdb macro support">). You'll have to substitute them
725yourself, or to invoke cpp on the source code files (see L</"The .i
726Targets">) So, for instance, you can't say
727
728 print SvPV_nolen(sv)
729
730but you have to say
731
732 print Perl_sv_2pv_nolen(sv)
733
734=back
735
736You may find it helpful to have a "macro dictionary", which you can
737produce by saying C<cpp -dM perl.c | sort>. Even then, F<cpp> won't
738recursively apply those macros for you.
739
740=head2 gdb macro support
741
742Recent versions of F<gdb> have fairly good macro support, but in order
743to use it you'll need to compile perl with macro definitions included
744in the debugging information. Using F<gcc> version 3.1, this means
745configuring with C<-Doptimize=-g3>. Other compilers might use a
746different switch (if they support debugging macros at all).
747
748=head2 Dumping Perl Data Structures
749
750One way to get around this macro hell is to use the dumping functions
751in F<dump.c>; these work a little like an internal
752L<Devel::Peek|Devel::Peek>, but they also cover OPs and other
753structures that you can't get at from Perl. Let's take an example.
754We'll use the C<$a = $b + $c> we used before, but give it a bit of
755context: C<$b = "6XXXX"; $c = 2.3;>. Where's a good place to stop and
756poke around?
757
758What about C<pp_add>, the function we examined earlier to implement the
759C<+> operator:
760
761 (gdb) break Perl_pp_add
762 Breakpoint 1 at 0x46249f: file pp_hot.c, line 309.
763
764Notice we use C<Perl_pp_add> and not C<pp_add> - see
765L<perlguts/Internal Functions>. With the breakpoint in place, we can
766run our program:
767
768 (gdb) run -e '$b = "6XXXX"; $c = 2.3; $a = $b + $c'
769
770Lots of junk will go past as gdb reads in the relevant source files and
771libraries, and then:
772
773 Breakpoint 1, Perl_pp_add () at pp_hot.c:309
774 309 dSP; dATARGET; tryAMAGICbin(add,opASSIGN);
775 (gdb) step
776 311 dPOPTOPnnrl_ul;
777 (gdb)
778
779We looked at this bit of code before, and we said that
780C<dPOPTOPnnrl_ul> arranges for two C<NV>s to be placed into C<left> and
781C<right> - let's slightly expand it:
782
783 #define dPOPTOPnnrl_ul NV right = POPn; \
784 SV *leftsv = TOPs; \
785 NV left = USE_LEFT(leftsv) ? SvNV(leftsv) : 0.0
786
787C<POPn> takes the SV from the top of the stack and obtains its NV
788either directly (if C<SvNOK> is set) or by calling the C<sv_2nv>
789function. C<TOPs> takes the next SV from the top of the stack - yes,
790C<POPn> uses C<TOPs> - but doesn't remove it. We then use C<SvNV> to
791get the NV from C<leftsv> in the same way as before - yes, C<POPn> uses
792C<SvNV>.
793
794Since we don't have an NV for C<$b>, we'll have to use C<sv_2nv> to
795convert it. If we step again, we'll find ourselves there:
796
797 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1669
798 1669 if (!sv)
799 (gdb)
800
801We can now use C<Perl_sv_dump> to investigate the SV:
802
803 SV = PV(0xa057cc0) at 0xa0675d0
804 REFCNT = 1
805 FLAGS = (POK,pPOK)
806 PV = 0xa06a510 "6XXXX"\0
807 CUR = 5
808 LEN = 6
809 $1 = void
810
811We know we're going to get C<6> from this, so let's finish the
812subroutine:
813
814 (gdb) finish
815 Run till exit from #0 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1671
816 0x462669 in Perl_pp_add () at pp_hot.c:311
817 311 dPOPTOPnnrl_ul;
818
819We can also dump out this op: the current op is always stored in
820C<PL_op>, and we can dump it with C<Perl_op_dump>. This'll give us
821similar output to L<B::Debug|B::Debug>.
822
823 {
824 13 TYPE = add ===> 14
825 TARG = 1
826 FLAGS = (SCALAR,KIDS)
827 {
828 TYPE = null ===> (12)
829 (was rv2sv)
830 FLAGS = (SCALAR,KIDS)
831 {
832 11 TYPE = gvsv ===> 12
833 FLAGS = (SCALAR)
834 GV = main::b
835 }
836 }
837
838# finish this later #
839
840=head1 SOURCE CODE STATIC ANALYSIS
841
842Various tools exist for analysing C source code B<statically>, as
843opposed to B<dynamically>, that is, without executing the code. It is
844possible to detect resource leaks, undefined behaviour, type
845mismatches, portability problems, code paths that would cause illegal
846memory accesses, and other similar problems by just parsing the C code
847and looking at the resulting graph, what does it tell about the
848execution and data flows. As a matter of fact, this is exactly how C
849compilers know to give warnings about dubious code.
850
851=head2 lint, splint
852
853The good old C code quality inspector, C<lint>, is available in several
854platforms, but please be aware that there are several different
855implementations of it by different vendors, which means that the flags
856are not identical across different platforms.
857
858There is a lint variant called C<splint> (Secure Programming Lint)
859available from http://www.splint.org/ that should compile on any
860Unix-like platform.
861
862There are C<lint> and <splint> targets in Makefile, but you may have to
863diddle with the flags (see above).
864
865=head2 Coverity
866
867Coverity (http://www.coverity.com/) is a product similar to lint and as
868a testbed for their product they periodically check several open source
869projects, and they give out accounts to open source developers to the
870defect databases.
871
872=head2 cpd (cut-and-paste detector)
873
874The cpd tool detects cut-and-paste coding. If one instance of the
875cut-and-pasted code changes, all the other spots should probably be
876changed, too. Therefore such code should probably be turned into a
877subroutine or a macro.
878
879cpd (http://pmd.sourceforge.net/cpd.html) is part of the pmd project
880(http://pmd.sourceforge.net/). pmd was originally written for static
881analysis of Java code, but later the cpd part of it was extended to
882parse also C and C++.
883
884Download the pmd-bin-X.Y.zip () from the SourceForge site, extract the
885pmd-X.Y.jar from it, and then run that on source code thusly:
886
0cbf2b31
FC
887 java -cp pmd-X.Y.jar net.sourceforge.pmd.cpd.CPD \
888 --minimum-tokens 100 --files /some/where/src --language c > cpd.txt
04c692a8
DR
889
890You may run into memory limits, in which case you should use the -Xmx
891option:
892
893 java -Xmx512M ...
894
895=head2 gcc warnings
896
897Though much can be written about the inconsistency and coverage
898problems of gcc warnings (like C<-Wall> not meaning "all the warnings",
899or some common portability problems not being covered by C<-Wall>, or
900C<-ansi> and C<-pedantic> both being a poorly defined collection of
901warnings, and so forth), gcc is still a useful tool in keeping our
902coding nose clean.
903
904The C<-Wall> is by default on.
905
906The C<-ansi> (and its sidekick, C<-pedantic>) would be nice to be on
907always, but unfortunately they are not safe on all platforms, they can
908for example cause fatal conflicts with the system headers (Solaris
909being a prime example). If Configure C<-Dgccansipedantic> is used, the
910C<cflags> frontend selects C<-ansi -pedantic> for the platforms where
911they are known to be safe.
912
913Starting from Perl 5.9.4 the following extra flags are added:
914
915=over 4
916
917=item *
918
919C<-Wendif-labels>
920
921=item *
922
923C<-Wextra>
924
925=item *
926
927C<-Wdeclaration-after-statement>
928
929=back
930
931The following flags would be nice to have but they would first need
932their own Augean stablemaster:
933
934=over 4
935
936=item *
937
938C<-Wpointer-arith>
939
940=item *
941
942C<-Wshadow>
943
944=item *
945
946C<-Wstrict-prototypes>
947
948=back
949
950The C<-Wtraditional> is another example of the annoying tendency of gcc
951to bundle a lot of warnings under one switch (it would be impossible to
952deploy in practice because it would complain a lot) but it does contain
953some warnings that would be beneficial to have available on their own,
954such as the warning about string constants inside macros containing the
955macro arguments: this behaved differently pre-ANSI than it does in
956ANSI, and some C compilers are still in transition, AIX being an
957example.
958
959=head2 Warnings of other C compilers
960
961Other C compilers (yes, there B<are> other C compilers than gcc) often
962have their "strict ANSI" or "strict ANSI with some portability
963extensions" modes on, like for example the Sun Workshop has its C<-Xa>
964mode on (though implicitly), or the DEC (these days, HP...) has its
965C<-std1> mode on.
966
967=head1 MEMORY DEBUGGERS
968
d1fd4856
VP
969B<NOTE 1>: Running under older memory debuggers such as Purify,
970valgrind or Third Degree greatly slows down the execution: seconds
971become minutes, minutes become hours. For example as of Perl 5.8.1, the
04c692a8
DR
972ext/Encode/t/Unicode.t takes extraordinarily long to complete under
973e.g. Purify, Third Degree, and valgrind. Under valgrind it takes more
974than six hours, even on a snappy computer. The said test must be doing
975something that is quite unfriendly for memory debuggers. If you don't
976feel like waiting, that you can simply kill away the perl process.
d1fd4856
VP
977Roughly valgrind slows down execution by factor 10, AddressSanitizer by
978factor 2.
04c692a8
DR
979
980B<NOTE 2>: To minimize the number of memory leak false alarms (see
981L</PERL_DESTRUCT_LEVEL> for more information), you have to set the
d8651d0d 982environment variable PERL_DESTRUCT_LEVEL to 2. For example, like this:
04c692a8
DR
983
984 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib ...
985
986B<NOTE 3>: There are known memory leaks when there are compile-time
987errors within eval or require, seeing C<S_doeval> in the call stack is
988a good sign of these. Fixing these leaks is non-trivial, unfortunately,
989but they must be fixed eventually.
990
991B<NOTE 4>: L<DynaLoader> will not clean up after itself completely
992unless Perl is built with the Configure option
993C<-Accflags=-DDL_UNLOAD_ALL_AT_EXIT>.
994
04c692a8
DR
995=head2 valgrind
996
d1fd4856
VP
997The valgrind tool can be used to find out both memory leaks and illegal
998heap memory accesses. As of version 3.3.0, Valgrind only supports Linux
999on x86, x86-64 and PowerPC and Darwin (OS X) on x86 and x86-64). The
1000special "test.valgrind" target can be used to run the tests under
1001valgrind. Found errors and memory leaks are logged in files named
1002F<testfile.valgrind>.
04c692a8
DR
1003
1004Valgrind also provides a cachegrind tool, invoked on perl as:
1005
1006 VG_OPTS=--tool=cachegrind make test.valgrind
1007
1008As system libraries (most notably glibc) are also triggering errors,
1009valgrind allows to suppress such errors using suppression files. The
1010default suppression file that comes with valgrind already catches a lot
1011of them. Some additional suppressions are defined in F<t/perl.supp>.
1012
1013To get valgrind and for more information see
1014
0061d4fa 1015 http://valgrind.org/
04c692a8 1016
81c3bbe7
RU
1017=head2 AddressSanitizer
1018
4dd56148
NC
1019AddressSanitizer is a clang and gcc extension, included in clang since
1020v3.1 and gcc since v4.8. It checks illegal heap pointers, global
1021pointers, stack pointers and use after free errors, and is fast enough
1022that you can easily compile your debugging or optimized perl with it.
1023It does not check memory leaks though. AddressSanitizer is available
1024for Linux, Mac OS X and soon on Windows.
81c3bbe7 1025
8a64fbaa
VP
1026To build perl with AddressSanitizer, your Configure invocation should
1027look like:
81c3bbe7 1028
e8596d90
VP
1029 sh Configure -des -Dcc=clang \
1030 -Accflags=-faddress-sanitizer -Aldflags=-faddress-sanitizer \
1031 -Alddlflags=-shared\ -faddress-sanitizer
81c3bbe7
RU
1032
1033where these arguments mean:
1034
1035=over 4
1036
1037=item * -Dcc=clang
1038
8a64fbaa
VP
1039This should be replaced by the full path to your clang executable if it
1040is not in your path.
81c3bbe7
RU
1041
1042=item * -Accflags=-faddress-sanitizer
1043
8a64fbaa 1044Compile perl and extensions sources with AddressSanitizer.
81c3bbe7
RU
1045
1046=item * -Aldflags=-faddress-sanitizer
1047
8a64fbaa 1048Link the perl executable with AddressSanitizer.
81c3bbe7 1049
e8596d90 1050=item * -Alddlflags=-shared\ -faddress-sanitizer
81c3bbe7 1051
e8596d90
VP
1052Link dynamic extensions with AddressSanitizer. You must manually
1053specify C<-shared> because using C<-Alddlflags=-shared> will prevent
1054Configure from setting a default value for C<lddlflags>, which usually
5dfc6e97 1055contains C<-shared> (at least on Linux).
81c3bbe7
RU
1056
1057=back
1058
8a64fbaa
VP
1059See also
1060L<http://code.google.com/p/address-sanitizer/wiki/AddressSanitizer>.
81c3bbe7
RU
1061
1062
04c692a8
DR
1063=head1 PROFILING
1064
1065Depending on your platform there are various ways of profiling Perl.
1066
1067There are two commonly used techniques of profiling executables:
1068I<statistical time-sampling> and I<basic-block counting>.
1069
1070The first method takes periodically samples of the CPU program counter,
1071and since the program counter can be correlated with the code generated
1072for functions, we get a statistical view of in which functions the
1073program is spending its time. The caveats are that very small/fast
1074functions have lower probability of showing up in the profile, and that
1075periodically interrupting the program (this is usually done rather
1076frequently, in the scale of milliseconds) imposes an additional
1077overhead that may skew the results. The first problem can be alleviated
1078by running the code for longer (in general this is a good idea for
1079profiling), the second problem is usually kept in guard by the
1080profiling tools themselves.
1081
1082The second method divides up the generated code into I<basic blocks>.
1083Basic blocks are sections of code that are entered only in the
1084beginning and exited only at the end. For example, a conditional jump
1085starts a basic block. Basic block profiling usually works by
1086I<instrumenting> the code by adding I<enter basic block #nnnn>
1087book-keeping code to the generated code. During the execution of the
1088code the basic block counters are then updated appropriately. The
1089caveat is that the added extra code can skew the results: again, the
1090profiling tools usually try to factor their own effects out of the
1091results.
1092
1093=head2 Gprof Profiling
1094
1095gprof is a profiling tool available in many Unix platforms, it uses
1096F<statistical time-sampling>.
1097
1098You can build a profiled version of perl called "perl.gprof" by
1099invoking the make target "perl.gprof" (What is required is that Perl
1100must be compiled using the C<-pg> flag, you may need to re-Configure).
1101Running the profiled version of Perl will create an output file called
1102F<gmon.out> is created which contains the profiling data collected
1103during the execution.
1104
1105The gprof tool can then display the collected data in various ways.
1106Usually gprof understands the following options:
1107
1108=over 4
1109
1110=item * -a
1111
1112Suppress statically defined functions from the profile.
1113
1114=item * -b
1115
1116Suppress the verbose descriptions in the profile.
1117
1118=item * -e routine
1119
1120Exclude the given routine and its descendants from the profile.
1121
1122=item * -f routine
1123
1124Display only the given routine and its descendants in the profile.
1125
1126=item * -s
1127
1128Generate a summary file called F<gmon.sum> which then may be given to
1129subsequent gprof runs to accumulate data over several runs.
1130
1131=item * -z
1132
1133Display routines that have zero usage.
1134
1135=back
1136
1137For more detailed explanation of the available commands and output
1138formats, see your own local documentation of gprof.
1139
1140quick hint:
1141
1142 $ sh Configure -des -Dusedevel -Doptimize='-pg' && make perl.gprof
1143 $ ./perl.gprof someprog # creates gmon.out in current directory
1144 $ gprof ./perl.gprof > out
1145 $ view out
1146
1147=head2 GCC gcov Profiling
1148
1149Starting from GCC 3.0 I<basic block profiling> is officially available
1150for the GNU CC.
1151
1152You can build a profiled version of perl called F<perl.gcov> by
1153invoking the make target "perl.gcov" (what is required that Perl must
1154be compiled using gcc with the flags C<-fprofile-arcs -ftest-coverage>,
1155you may need to re-Configure).
1156
1157Running the profiled version of Perl will cause profile output to be
1158generated. For each source file an accompanying ".da" file will be
1159created.
1160
1161To display the results you use the "gcov" utility (which should be
1162installed if you have gcc 3.0 or newer installed). F<gcov> is run on
1163source code files, like this
1164
1165 gcov sv.c
1166
1167which will cause F<sv.c.gcov> to be created. The F<.gcov> files contain
1168the source code annotated with relative frequencies of execution
1169indicated by "#" markers.
1170
1171Useful options of F<gcov> include C<-b> which will summarise the basic
1172block, branch, and function call coverage, and C<-c> which instead of
1173relative frequencies will use the actual counts. For more information
1174on the use of F<gcov> and basic block profiling with gcc, see the
1175latest GNU CC manual, as of GCC 3.0 see
1176
1177 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc.html
1178
1179and its section titled "8. gcov: a Test Coverage Program"
1180
1181 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc_8.html#SEC132
1182
1183quick hint:
1184
1185 $ sh Configure -des -Dusedevel -Doptimize='-g' \
1186 -Accflags='-fprofile-arcs -ftest-coverage' \
1187 -Aldflags='-fprofile-arcs -ftest-coverage' && make perl.gcov
1188 $ rm -f regexec.c.gcov regexec.gcda
1189 $ ./perl.gcov
1190 $ gcov regexec.c
1191 $ view regexec.c.gcov
1192
1193=head1 MISCELLANEOUS TRICKS
1194
1195=head2 PERL_DESTRUCT_LEVEL
1196
1197If you want to run any of the tests yourself manually using e.g.
4dd56148
NC
1198valgrind, please note that by default perl B<does not> explicitly
1199cleanup all the memory it has allocated (such as global memory arenas)
1200but instead lets the exit() of the whole program "take care" of such
1201allocations, also known as "global destruction of objects".
04c692a8
DR
1202
1203There is a way to tell perl to do complete cleanup: set the environment
1204variable PERL_DESTRUCT_LEVEL to a non-zero value. The t/TEST wrapper
1205does set this to 2, and this is what you need to do too, if you don't
f01ecde8 1206want to see the "global leaks": For example, for running under valgrind
04c692a8 1207
f01ecde8 1208 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib t/foo/bar.t
04c692a8
DR
1209
1210(Note: the mod_perl apache module uses also this environment variable
1211for its own purposes and extended its semantics. Refer to the mod_perl
1212documentation for more information. Also, spawned threads do the
1213equivalent of setting this variable to the value 1.)
1214
1215If, at the end of a run you get the message I<N scalars leaked>, you
1216can recompile with C<-DDEBUG_LEAKING_SCALARS>, which will cause the
1217addresses of all those leaked SVs to be dumped along with details as to
1218where each SV was originally allocated. This information is also
1219displayed by Devel::Peek. Note that the extra details recorded with
1220each SV increases memory usage, so it shouldn't be used in production
1221environments. It also converts C<new_SV()> from a macro into a real
1222function, so you can use your favourite debugger to discover where
1223those pesky SVs were allocated.
1224
1225If you see that you're leaking memory at runtime, but neither valgrind
1226nor C<-DDEBUG_LEAKING_SCALARS> will find anything, you're probably
1227leaking SVs that are still reachable and will be properly cleaned up
1228during destruction of the interpreter. In such cases, using the C<-Dm>
1229switch can point you to the source of the leak. If the executable was
1230built with C<-DDEBUG_LEAKING_SCALARS>, C<-Dm> will output SV
1231allocations in addition to memory allocations. Each SV allocation has a
1232distinct serial number that will be written on creation and destruction
1233of the SV. So if you're executing the leaking code in a loop, you need
1234to look for SVs that are created, but never destroyed between each
1235cycle. If such an SV is found, set a conditional breakpoint within
1236C<new_SV()> and make it break only when C<PL_sv_serial> is equal to the
1237serial number of the leaking SV. Then you will catch the interpreter in
1238exactly the state where the leaking SV is allocated, which is
1239sufficient in many cases to find the source of the leak.
1240
1241As C<-Dm> is using the PerlIO layer for output, it will by itself
1242allocate quite a bunch of SVs, which are hidden to avoid recursion. You
1243can bypass the PerlIO layer if you use the SV logging provided by
1244C<-DPERL_MEM_LOG> instead.
1245
1246=head2 PERL_MEM_LOG
1247
1248If compiled with C<-DPERL_MEM_LOG>, both memory and SV allocations go
1249through logging functions, which is handy for breakpoint setting.
1250
1251Unless C<-DPERL_MEM_LOG_NOIMPL> is also compiled, the logging functions
1252read $ENV{PERL_MEM_LOG} to determine whether to log the event, and if
1253so how:
1254
1255 $ENV{PERL_MEM_LOG} =~ /m/ Log all memory ops
1256 $ENV{PERL_MEM_LOG} =~ /s/ Log all SV ops
1257 $ENV{PERL_MEM_LOG} =~ /t/ include timestamp in Log
1258 $ENV{PERL_MEM_LOG} =~ /^(\d+)/ write to FD given (default is 2)
1259
1260Memory logging is somewhat similar to C<-Dm> but is independent of
1261C<-DDEBUGGING>, and at a higher level; all uses of Newx(), Renew(), and
1262Safefree() are logged with the caller's source code file and line
1263number (and C function name, if supported by the C compiler). In
1264contrast, C<-Dm> is directly at the point of C<malloc()>. SV logging is
1265similar.
1266
1267Since the logging doesn't use PerlIO, all SV allocations are logged and
1268no extra SV allocations are introduced by enabling the logging. If
1269compiled with C<-DDEBUG_LEAKING_SCALARS>, the serial number for each SV
1270allocation is also logged.
1271
1272=head2 DDD over gdb
1273
1274Those debugging perl with the DDD frontend over gdb may find the
1275following useful:
1276
1277You can extend the data conversion shortcuts menu, so for example you
1278can display an SV's IV value with one click, without doing any typing.
1279To do that simply edit ~/.ddd/init file and add after:
1280
1281 ! Display shortcuts.
1282 Ddd*gdbDisplayShortcuts: \
1283 /t () // Convert to Bin\n\
1284 /d () // Convert to Dec\n\
1285 /x () // Convert to Hex\n\
1286 /o () // Convert to Oct(\n\
1287
1288the following two lines:
1289
1290 ((XPV*) (())->sv_any )->xpv_pv // 2pvx\n\
1291 ((XPVIV*) (())->sv_any )->xiv_iv // 2ivx
1292
1293so now you can do ivx and pvx lookups or you can plug there the sv_peek
1294"conversion":
1295
1296 Perl_sv_peek(my_perl, (SV*)()) // sv_peek
1297
1298(The my_perl is for threaded builds.) Just remember that every line,
1299but the last one, should end with \n\
1300
1301Alternatively edit the init file interactively via: 3rd mouse button ->
1302New Display -> Edit Menu
1303
1304Note: you can define up to 20 conversion shortcuts in the gdb section.
1305
1306=head2 Poison
1307
1308If you see in a debugger a memory area mysteriously full of 0xABABABAB
1309or 0xEFEFEFEF, you may be seeing the effect of the Poison() macros, see
1310L<perlclib>.
1311
1312=head2 Read-only optrees
1313
1314Under ithreads the optree is read only. If you want to enforce this, to
1315check for write accesses from buggy code, compile with
04c692a8 1316C<-DPERL_DEBUG_READONLY_OPS> to enable code that allocates op memory
4dd56148
NC
1317via C<mmap>, and sets it read-only when it is attached to a subroutine.
1318Any write access to an op results in a C<SIGBUS> and abort.
04c692a8
DR
1319
1320This code is intended for development only, and may not be portable
1321even to all Unix variants. Also, it is an 80% solution, in that it
4dd56148
NC
1322isn't able to make all ops read only. Specifically it does not apply to
1323op slabs belonging to C<BEGIN> blocks.
04c692a8 1324
4dd56148
NC
1325However, as an 80% solution it is still effective, as it has caught
1326bugs in the past.
04c692a8
DR
1327
1328=head2 The .i Targets
1329
1330You can expand the macros in a F<foo.c> file by saying
1331
1332 make foo.i
1333
d1fd4856
VP
1334which will expand the macros using cpp. Don't be scared by the
1335results.
04c692a8
DR
1336
1337=head1 AUTHOR
1338
1339This document was originally written by Nathan Torkington, and is
1340maintained by the perl5-porters mailing list.