1PERLHACKTIPS(1) Perl Programmers Reference Guide PERLHACKTIPS(1)
2
6 perlhacktips - Tips for Perl core C code hacking
7
9 This document will help you learn the best way to go about hacking on
10 the Perl core C code. It covers common problems, debugging, profiling,
11 and more.
12 If you haven't read perlhack and perlhacktut yet, you might want to do
14 that first.
15
17 Perl source now permits some specific C99 features which we know are
18 supported by all platforms, but mostly plays by ANSI C89 rules. You
19 don't care about some particular platform having broken Perl? I hear
20 there is still a strong demand for J2EE programmers.
21 Perl environment problems
23 • Not compiling with threading
24 Compiling with threading (-Duseithreads) completely rewrites the
26 function prototypes of Perl. You better try your changes with
27 that. Related to this is the difference between "Perl_-less" and
28 "Perl_-ly" APIs, for example:
29 Perl_sv_setiv(aTHX_ ...);
31 sv_setiv(...);
32 The first one explicitly passes in the context, which is needed for
34 e.g. threaded builds. The second one does that implicitly; do not
35 get them mixed. If you are not passing in a aTHX_, you will need
36 to do a dTHX as the first thing in the function.
37 See "How multiple interpreters and concurrency are supported" in
39 perlguts for further discussion about context.
40 • Not compiling with -DDEBUGGING
42 The DEBUGGING define exposes more code to the compiler, therefore
44 more ways for things to go wrong. You should try it.
45 • Introducing (non-read-only) globals
47 Do not introduce any modifiable globals, truly global or file
49 static. They are bad form and complicate multithreading and other
50 forms of concurrency. The right way is to introduce them as new
51 interpreter variables, see intrpvar.h (at the very end for binary
52 compatibility).
53 Introducing read-only (const) globals is okay, as long as you
55 verify with e.g. "nm libperl.a|egrep -v ' [TURtr] '" (if your "nm"
56 has BSD-style output) that the data you added really is read-only.
57 (If it is, it shouldn't show up in the output of that command.)
58 If you want to have static strings, make them constant:
60 static const char etc[] = "...";
62 If you want to have arrays of constant strings, note carefully the
64 right combination of "const"s:
65 static const char * const yippee[] =
67 {"hi", "ho", "silver"};
68 • Not exporting your new function
70 Some platforms (Win32, AIX, VMS, OS/2, to name a few) require any
72 function that is part of the public API (the shared Perl library)
73 to be explicitly marked as exported. See the discussion about
74 embed.pl in perlguts.
75 • Exporting your new function
77 The new shiny result of either genuine new functionality or your
79 arduous refactoring is now ready and correctly exported. So what
80 could possibly go wrong?
81 Maybe simply that your function did not need to be exported in the
83 first place. Perl has a long and not so glorious history of
84 exporting functions that it should not have.
85 If the function is used only inside one source code file, make it
87 static. See the discussion about embed.pl in perlguts.
88 If the function is used across several files, but intended only for
90 Perl's internal use (and this should be the common case), do not
91 export it to the public API. See the discussion about embed.pl in
92 perlguts.
93 C99
95 Starting from 5.35.5 we now permit some C99 features in the core C
96 source. However, code in dual life extensions still needs to be C89
97 only, because it needs to compile against earlier version of Perl
98 running on older platforms. Also note that our headers need to also be
99 valid as C++, because XS extensions written in C++ need to include
100 them, hence member structure initialisers can't be used in headers.
101 C99 support is still far from complete on all platforms we currently
103 support. As a baseline we can only assume C89 semantics with the
104 specific C99 features described below, which we've verified work
105 everywhere. It's fine to probe for additional C99 features and use
106 them where available, providing there is also a fallback for compilers
107 that don't support the feature. For example, we use C11 thread local
108 storage when available, but fall back to POSIX thread specific APIs
109 otherwise, and we use "char" for booleans if "<stdbool.h>" isn't
110 available.
111 Code can use (and rely on) the following C99 features being present
113 • mixed declarations and code
115 • 64 bit integer types
117 For consistency with the existing source code, use the typedefs
119 "I64" and "U64", instead of using "long long" and "unsigned long
120 long" directly.
121 • variadic macros
123 void greet(char *file, unsigned int line, char *format, ...);
125 #define logged_greet(...) greet(__FILE__, __LINE__, __VA_ARGS__);
126 Note that "__VA_OPT__" is a gcc extension not yet in any published
128 standard.
129 • declarations in for loops
131 for (const char *p = message; *p; ++p) {
133 putchar(*p);
134 }
135 • member structure initialisers
137 But not in headers, as support was only added to C++ relatively
139 recently.
140 Hence this is fine in C and XS code, but not headers:
142 struct message {
144 char *action;
145 char *target;
146 };
147 struct message mcguffin = {
149 .target = "member structure initialisers",
150 .action = "Built"
151 };
152 • flexible array members
154 This is standards conformant:
156 struct greeting {
158 unsigned int len;
159 char message[];
160 };
161 However, the source code already uses the "unwarranted chumminess
163 with the compiler" hack in many places:
164 struct greeting {
166 unsigned int len;
167 char message[1];
168 };
169 Strictly it is undefined behaviour accessing beyond "message[0]",
171 but this has been a commonly used hack since K&R times, and using
172 it hasn't been a practical issue anywhere (in the perl source or
173 any other common C code). Hence it's unclear what we would gain
174 from actively changing to the C99 approach.
175 • "//" comments
177 All compilers we tested support their use. Not all humans we tested
179 support their use.
180 Code explicitly should not use any other C99 features. For example
182 • variable length arrays
184 Not supported by any MSVC, and this is not going to change.
186 Even "variable" length arrays where the variable is a constant
188 expression are syntax errors under MSVC.
189 • C99 types in "<stdint.h>"
191 Use "PERL_INT_FAST8_T" etc as defined in handy.h
193 • C99 format strings in "<inttypes.h>"
195 "snprintf" in the VMS libc only added support for "PRIdN" etc very
197 recently, meaning that there are live supported installations
198 without this, or formats such as %zu.
199 (perl's "sv_catpvf" etc use parser code code in "sv.c", which
201 supports the "z" modifier, along with perl-specific formats such as
202 "SVf".)
203 If you want to use a C99 feature not listed above then you need to do
205 one of
206 • Probe for it in Configure, set a variable in config.sh, and add
208 fallback logic in the headers for platforms which don't have it.
209 • Write test code and verify that it works on platforms we need to
211 support, before relying on it unconditionally.
212 Likely you want to repeat the same plan as we used to get the current
214 C99 feature set. See the message at
215 https://markmail.org/thread/odr4fjrn72u2fkpz for the C99 probes we used
216 before. Note that the two most "fussy" compilers appear to be MSVC and
217 the vendor compiler on VMS. To date all the *nix compilers have been
218 far more flexible in what they support.
219 On *nix platforms, Configure attempts to set compiler flags
221 appropriately. All vendor compilers that we tested defaulted to C99
222 (or C11) support. However, older versions of gcc default to C89, or
223 permit most C99 (with warnings), but forbid declarations in for loops
224 unless "-std=gnu99" is added. The alternative "-std=c99" might seem
225 better, but using it on some platforms can prevent "<unistd.h>"
226 declaring some prototypes being declared, which breaks the build. gcc's
227 "-ansi" flag implies "-std=c89" so we can no longer set that, hence the
228 Configure option "-gccansipedantic" now only adds "-pedantic".
229 The Perl core source code files (the ones at the top level of the
231 source code distribution) are automatically compiled with as many as
232 possible of the "-std=gnu99", "-pedantic", and a selection of "-W"
233 flags (see cflags.SH). Files in ext/ dist/ cpan/ etc are compiled with
234 the same flags as the installed perl would use to compile XS
235 extensions.
236 Basically, it's safe to assume that Configure and cflags.SH have picked
238 the best combination of flags for the version of gcc on the platform,
239 and attempting to add more flags related to enforcing a C dialect will
240 cause problems either locally, or on other systems that the code is
241 shipped to.
242 We believe that the C99 support in gcc 3.1 is good enough for us, but
244 we don't have a 19 year old gcc handy to check this :-) If you have
245 ancient vendor compilers that don't default to C99, the flags you might
246 want to try are
247 AIX "-qlanglvl=stdc99"
249 HP/UX
251 "-AC99"
252 Solaris
254 "-xc99"
255 Portability problems
257 The following are common causes of compilation and/or execution
258 failures, not common to Perl as such. The C FAQ is good bedtime
259 reading. Please test your changes with as many C compilers and
260 platforms as possible; we will, anyway, and it's nice to save oneself
261 from public embarrassment.
262 Also study perlport carefully to avoid any bad assumptions about the
264 operating system, filesystems, character set, and so forth.
265 Do not assume an operating system indicates a certain compiler.
267 • Casting pointers to integers or casting integers to pointers
269 void castaway(U8* p)
271 {
272 IV i = p;
273 or
275 void castaway(U8* p)
277 {
278 IV i = (IV)p;
279 Both are bad, and broken, and unportable. Use the PTR2IV() macro
281 that does it right. (Likewise, there are PTR2UV(), PTR2NV(),
282 INT2PTR(), and NUM2PTR().)
283 • Casting between function pointers and data pointers
285 Technically speaking casting between function pointers and data
287 pointers is unportable and undefined, but practically speaking it
288 seems to work, but you should use the FPTR2DPTR() and DPTR2FPTR()
289 macros. Sometimes you can also play games with unions.
290 • Assuming sizeof(int) == sizeof(long)
292 There are platforms where longs are 64 bits, and platforms where
294 ints are 64 bits, and while we are out to shock you, even platforms
295 where shorts are 64 bits. This is all legal according to the C
296 standard. (In other words, "long long" is not a portable way to
297 specify 64 bits, and "long long" is not even guaranteed to be any
298 wider than "long".)
299 Instead, use the definitions IV, UV, IVSIZE, I32SIZE, and so forth.
301 Avoid things like I32 because they are not guaranteed to be exactly
302 32 bits, they are at least 32 bits, nor are they guaranteed to be
303 int or long. If you explicitly need 64-bit variables, use I64 and
304 U64.
305 • Assuming one can dereference any type of pointer for any type of
307 data
308 char *p = ...;
310 long pony = *(long *)p; /* BAD */
311 Many platforms, quite rightly so, will give you a core dump instead
313 of a pony if the p happens not to be correctly aligned.
314 • Lvalue casts
316 (int)*p = ...; /* BAD */
318 Simply not portable. Get your lvalue to be of the right type, or
320 maybe use temporary variables, or dirty tricks with unions.
321 • Assume anything about structs (especially the ones you don't
323 control, like the ones coming from the system headers)
324 • That a certain field exists in a struct
326 • That no other fields exist besides the ones you know of
328 • That a field is of certain signedness, sizeof, or type
330 • That the fields are in a certain order
332 • While C guarantees the ordering specified in the
334 struct definition, between different platforms the
335 definitions might differ
336 • That the sizeof(struct) or the alignments are the same
338 everywhere
339 • There might be padding bytes between the fields to
341 align the fields - the bytes can be anything
342 • Structs are required to be aligned to the maximum
344 alignment required by the fields - which for native
345 types is for usually equivalent to sizeof() of the
346 field
347 • Assuming the character set is ASCIIish
349 Perl can compile and run under EBCDIC platforms. See perlebcdic.
351 This is transparent for the most part, but because the character
352 sets differ, you shouldn't use numeric (decimal, octal, nor hex)
353 constants to refer to characters. You can safely say 'A', but not
354 0x41. You can safely say '\n', but not "\012". However, you can
355 use macros defined in utf8.h to specify any code point portably.
356 "LATIN1_TO_NATIVE(0xDF)" is going to be the code point that means
357 LATIN SMALL LETTER SHARP S on whatever platform you are running on
358 (on ASCII platforms it compiles without adding any extra code, so
359 there is zero performance hit on those). The acceptable inputs to
360 "LATIN1_TO_NATIVE" are from 0x00 through 0xFF. If your input isn't
361 guaranteed to be in that range, use "UNICODE_TO_NATIVE" instead.
362 "NATIVE_TO_LATIN1" and "NATIVE_TO_UNICODE" translate the opposite
363 direction.
364 If you need the string representation of a character that doesn't
366 have a mnemonic name in C, you should add it to the list in
367 regen/unicode_constants.pl, and have Perl create "#define"'s for
368 you, based on the current platform.
369 Note that the "isFOO" and "toFOO" macros in handy.h work properly
371 on native code points and strings.
372 Also, the range 'A' - 'Z' in ASCII is an unbroken sequence of 26
374 upper case alphabetic characters. That is not true in EBCDIC. Nor
375 for 'a' to 'z'. But '0' - '9' is an unbroken range in both
376 systems. Don't assume anything about other ranges. (Note that
377 special handling of ranges in regular expression patterns and
378 transliterations makes it appear to Perl code that the
379 aforementioned ranges are all unbroken.)
380 Many of the comments in the existing code ignore the possibility of
382 EBCDIC, and may be wrong therefore, even if the code works. This
383 is actually a tribute to the successful transparent insertion of
384 being able to handle EBCDIC without having to change pre-existing
385 code.
386 UTF-8 and UTF-EBCDIC are two different encodings used to represent
388 Unicode code points as sequences of bytes. Macros with the same
389 names (but different definitions) in utf8.h and utfebcdic.h are
390 used to allow the calling code to think that there is only one such
391 encoding. This is almost always referred to as "utf8", but it
392 means the EBCDIC version as well. Again, comments in the code may
393 well be wrong even if the code itself is right. For example, the
394 concept of UTF-8 "invariant characters" differs between ASCII and
395 EBCDIC. On ASCII platforms, only characters that do not have the
396 high-order bit set (i.e. whose ordinals are strict ASCII, 0 - 127)
397 are invariant, and the documentation and comments in the code may
398 assume that, often referring to something like, say, "hibit". The
399 situation differs and is not so simple on EBCDIC machines, but as
400 long as the code itself uses the "NATIVE_IS_INVARIANT()" macro
401 appropriately, it works, even if the comments are wrong.
402 As noted in "TESTING" in perlhack, when writing test scripts, the
404 file t/charset_tools.pl contains some helpful functions for writing
405 tests valid on both ASCII and EBCDIC platforms. Sometimes, though,
406 a test can't use a function and it's inconvenient to have different
407 test versions depending on the platform. There are 20 code points
408 that are the same in all 4 character sets currently recognized by
409 Perl (the 3 EBCDIC code pages plus ISO 8859-1 (ASCII/Latin1)).
410 These can be used in such tests, though there is a small
411 possibility that Perl will become available in yet another
412 character set, breaking your test. All but one of these code
413 points are C0 control characters. The most significant controls
414 that are the same are "\0", "\r", and "\N{VT}" (also specifiable as
415 "\cK", "\x0B", "\N{U+0B}", or "\013"). The single non-control is
416 U+00B6 PILCROW SIGN. The controls that are the same have the same
417 bit pattern in all 4 character sets, regardless of the UTF8ness of
418 the string containing them. The bit pattern for U+B6 is the same
419 in all 4 for non-UTF8 strings, but differs in each when its
420 containing string is UTF-8 encoded. The only other code points
421 that have some sort of sameness across all 4 character sets are the
422 pair 0xDC and 0xFC. Together these represent upper- and lowercase
423 LATIN LETTER U WITH DIAERESIS, but which is upper and which is
424 lower may be reversed: 0xDC is the capital in Latin1 and 0xFC is
425 the small letter, while 0xFC is the capital in EBCDIC and 0xDC is
426 the small one. This factoid may be exploited in writing case
427 insensitive tests that are the same across all 4 character sets.
428 • Assuming the character set is just ASCII
430 ASCII is a 7 bit encoding, but bytes have 8 bits in them. The 128
432 extra characters have different meanings depending on the locale.
433 Absent a locale, currently these extra characters are generally
434 considered to be unassigned, and this has presented some problems.
435 This has being changed starting in 5.12 so that these characters
436 can be considered to be Latin-1 (ISO-8859-1).
437 • Mixing #define and #ifdef
439 #define BURGLE(x) ... \
441 #ifdef BURGLE_OLD_STYLE /* BAD */
442 ... do it the old way ... \
443 #else
444 ... do it the new way ... \
445 #endif
446 You cannot portably "stack" cpp directives. For example in the
448 above you need two separate BURGLE() #defines, one for each #ifdef
449 branch.
450 • Adding non-comment stuff after #endif or #else
452 #ifdef SNOSH
454 ...
455 #else !SNOSH /* BAD */
456 ...
457 #endif SNOSH /* BAD */
458 The #endif and #else cannot portably have anything non-comment
460 after them. If you want to document what is going (which is a good
461 idea especially if the branches are long), use (C) comments:
462 #ifdef SNOSH
464 ...
465 #else /* !SNOSH */
466 ...
467 #endif /* SNOSH */
468 The gcc option "-Wendif-labels" warns about the bad variant (by
470 default on starting from Perl 5.9.4).
471 • Having a comma after the last element of an enum list
473 enum color {
475 CERULEAN,
476 CHARTREUSE,
477 CINNABAR, /* BAD */
478 };
479 is not portable. Leave out the last comma.
481 Also note that whether enums are implicitly morphable to ints
483 varies between compilers, you might need to (int).
484 • Mixing signed char pointers with unsigned char pointers
486 int foo(char *s) { ... }
488 ...
489 unsigned char *t = ...; /* Or U8* t = ... */
490 foo(t); /* BAD */
491 While this is legal practice, it is certainly dubious, and
493 downright fatal in at least one platform: for example VMS cc
494 considers this a fatal error. One cause for people often making
495 this mistake is that a "naked char" and therefore dereferencing a
496 "naked char pointer" have an undefined signedness: it depends on
497 the compiler and the flags of the compiler and the underlying
498 platform whether the result is signed or unsigned. For this very
499 same reason using a 'char' as an array index is bad.
500 • Macros that have string constants and their arguments as substrings
502 of the string constants
503 #define FOO(n) printf("number = %d\n", n) /* BAD */
505 FOO(10);
506 Pre-ANSI semantics for that was equivalent to
508 printf("10umber = %d\10");
510 which is probably not what you were expecting. Unfortunately at
512 least one reasonably common and modern C compiler does "real
513 backward compatibility" here, in AIX that is what still happens
514 even though the rest of the AIX compiler is very happily C89.
515 • Using printf formats for non-basic C types
517 IV i = ...;
519 printf("i = %d\n", i); /* BAD */
520 While this might by accident work in some platform (where IV
522 happens to be an "int"), in general it cannot. IV might be
523 something larger. Even worse the situation is with more specific
524 types (defined by Perl's configuration step in config.h):
525 Uid_t who = ...;
527 printf("who = %d\n", who); /* BAD */
528 The problem here is that Uid_t might be not only not "int"-wide but
530 it might also be unsigned, in which case large uids would be
531 printed as negative values.
532 There is no simple solution to this because of printf()'s limited
534 intelligence, but for many types the right format is available as
535 with either 'f' or '_f' suffix, for example:
536 IVdf /* IV in decimal */
538 UVxf /* UV is hexadecimal */
539 printf("i = %"IVdf"\n", i); /* The IVdf is a string constant. */
541 Uid_t_f /* Uid_t in decimal */
543 printf("who = %"Uid_t_f"\n", who);
545 Or you can try casting to a "wide enough" type:
547 printf("i = %"IVdf"\n", (IV)something_very_small_and_signed);
549 See "Formatted Printing of Size_t and SSize_t" in perlguts for how
551 to print those.
552 Also remember that the %p format really does require a void
554 pointer:
555 U8* p = ...;
557 printf("p = %p\n", (void*)p);
558 The gcc option "-Wformat" scans for such problems.
560 • Blindly passing va_list
562 Not all platforms support passing va_list to further varargs
564 (stdarg) functions. The right thing to do is to copy the va_list
565 using the Perl_va_copy() if the NEED_VA_COPY is defined.
566 • Using gcc statement expressions
568 val = ({...;...;...}); /* BAD */
570 While a nice extension, it's not portable. Historically, Perl used
572 them in macros if available to gain some extra speed (essentially
573 as a funky form of inlining), but we now support (or emulate) C99
574 "static inline" functions, so use them instead. Declare functions
575 as "PERL_STATIC_INLINE" to transparently fall back to emulation
576 where needed.
577 • Binding together several statements in a macro
579 Use the macros STMT_START and STMT_END.
581 STMT_START {
583 ...
584 } STMT_END
585 • Testing for operating systems or versions when should be testing
587 for features
588 #ifdef __FOONIX__ /* BAD */
590 foo = quux();
591 #endif
592 Unless you know with 100% certainty that quux() is only ever
594 available for the "Foonix" operating system and that is available
595 and correctly working for all past, present, and future versions of
596 "Foonix", the above is very wrong. This is more correct (though
597 still not perfect, because the below is a compile-time check):
598 #ifdef HAS_QUUX
600 foo = quux();
601 #endif
602 How does the HAS_QUUX become defined where it needs to be? Well,
604 if Foonix happens to be Unixy enough to be able to run the
605 Configure script, and Configure has been taught about detecting and
606 testing quux(), the HAS_QUUX will be correctly defined. In other
607 platforms, the corresponding configuration step will hopefully do
608 the same.
609 In a pinch, if you cannot wait for Configure to be educated, or if
611 you have a good hunch of where quux() might be available, you can
612 temporarily try the following:
613 #if (defined(__FOONIX__) || defined(__BARNIX__))
615 # define HAS_QUUX
616 #endif
617 ...
619 #ifdef HAS_QUUX
621 foo = quux();
622 #endif
623 But in any case, try to keep the features and operating systems
625 separate.
626 A good resource on the predefined macros for various operating
628 systems, compilers, and so forth is
629 <http://sourceforge.net/p/predef/wiki/Home/>
630 • Assuming the contents of static memory pointed to by the return
632 values of Perl wrappers for C library functions doesn't change.
633 Many C library functions return pointers to static storage that can
634 be overwritten by subsequent calls to the same or related
635 functions. Perl has wrappers for some of these functions.
636 Originally many of those wrappers returned those volatile pointers.
637 But over time almost all of them have evolved to return stable
638 copies. To cope with the remaining ones, do a "savepv" in perlapi
639 to make a copy, thus avoiding these problems. You will have to
640 free the copy when you're done to avoid memory leaks. If you don't
641 have control over when it gets freed, you'll need to make the copy
642 in a mortal scalar, like so
643 SvPVX(sv_2mortal(newSVpv(volatile_string, 0)))
645 Problematic System Interfaces
647 • Perl strings are NOT the same as C strings: They may contain "NUL"
648 characters, whereas a C string is terminated by the first "NUL".
649 That is why Perl API functions that deal with strings generally
650 take a pointer to the first byte and either a length or a pointer
651 to the byte just beyond the final one.
652 And this is the reason that many of the C library string handling
654 functions should not be used. They don't cope with the full
655 generality of Perl strings. It may be that your test cases don't
656 have embedded "NUL"s, and so the tests pass, whereas there may well
657 eventually arise real-world cases where they fail. A lesson here
658 is to include "NUL"s in your tests. Now it's fairly rare in most
659 real world cases to get "NUL"s, so your code may seem to work,
660 until one day a "NUL" comes along.
661 Here's an example. It used to be a common paradigm, for decades,
663 in the perl core to use "strchr("list", c)" to see if the character
664 "c" is any of the ones given in "list", a double-quote-enclosed
665 string of the set of characters that we are seeing if "c" is one
666 of. As long as "c" isn't a "NUL", it works. But when "c" is a
667 "NUL", "strchr" returns a pointer to the terminating "NUL" in
668 "list". This likely will result in a segfault or a security issue
669 when the caller uses that end pointer as the starting point to read
670 from.
671 A solution to this and many similar issues is to use the "mem"-foo
673 C library functions instead. In this case "memchr" can be used to
674 see if "c" is in "list" and works even if "c" is "NUL". These
675 functions need an additional parameter to give the string length.
676 In the case of literal string parameters, perl has defined macros
677 that calculate the length for you. See "String Handling" in
678 perlapi.
679 • malloc(0), realloc(0), calloc(0, 0) are non-portable. To be
681 portable allocate at least one byte. (In general you should rarely
682 need to work at this low level, but instead use the various malloc
683 wrappers.)
684 • snprintf() - the return type is unportable. Use my_snprintf()
686 instead.
687 Security problems
689 Last but not least, here are various tips for safer coding. See also
690 perlclib for libc/stdio replacements one should use.
691 • Do not use gets()
693 Or we will publicly ridicule you. Seriously.
695 • Do not use tmpfile()
697 Use mkstemp() instead.
699 • Do not use strcpy() or strcat() or strncpy() or strncat()
701 Use my_strlcpy() and my_strlcat() instead: they either use the
703 native implementation, or Perl's own implementation (borrowed from
704 the public domain implementation of INN).
705 • Do not use sprintf() or vsprintf()
707 If you really want just plain byte strings, use my_snprintf() and
709 my_vsnprintf() instead, which will try to use snprintf() and
710 vsnprintf() if those safer APIs are available. If you want
711 something fancier than a plain byte string, use "Perl_form"() or
712 SVs and "Perl_sv_catpvf()".
713 Note that glibc "printf()", "sprintf()", etc. are buggy before
715 glibc version 2.17. They won't allow a "%.s" format with a
716 precision to create a string that isn't valid UTF-8 if the current
717 underlying locale of the program is UTF-8. What happens is that
718 the %s and its operand are simply skipped without any notice.
719 <https://sourceware.org/bugzilla/show_bug.cgi?id=6530>.
720 • Do not use atoi()
722 Use grok_atoUV() instead. atoi() has ill-defined behavior on
724 overflows, and cannot be used for incremental parsing. It is also
725 affected by locale, which is bad.
726 • Do not use strtol() or strtoul()
728 Use grok_atoUV() instead. strtol() or strtoul() (or their
730 IV/UV-friendly macro disguises, Strtol() and Strtoul(), or Atol()
731 and Atoul() are affected by locale, which is bad.
732
734 You can compile a special debugging version of Perl, which allows you
735 to use the "-D" option of Perl to tell more about what Perl is doing.
736 But sometimes there is no alternative than to dive in with a debugger,
737 either to see the stack trace of a core dump (very useful in a bug
738 report), or trying to figure out what went wrong before the core dump
739 happened, or how did we end up having wrong or unexpected results.
740 Poking at Perl
742 To really poke around with Perl, you'll probably want to build Perl for
743 debugging, like this:
744 ./Configure -d -DDEBUGGING
746 make
747 "-DDEBUGGING" turns on the C compiler's "-g" flag to have it produce
749 debugging information which will allow us to step through a running
750 program, and to see in which C function we are at (without the
751 debugging information we might see only the numerical addresses of the
752 functions, which is not very helpful). It will also turn on the
753 "DEBUGGING" compilation symbol which enables all the internal debugging
754 code in Perl. There are a whole bunch of things you can debug with
755 this: perlrun lists them all, and the best way to find out about them
756 is to play about with them. The most useful options are probably
757 l Context (loop) stack processing
759 s Stack snapshots (with v, displays all stacks)
760 t Trace execution
761 o Method and overloading resolution
762 c String/numeric conversions
763 For example
765 $ perl -Dst -e '$a + 1'
767 ....
768 (-e:1) gvsv(main::a)
769 => UNDEF
770 (-e:1) const(IV(1))
771 => UNDEF IV(1)
772 (-e:1) add
773 => NV(1)
774 Some of the functionality of the debugging code can be achieved with a
776 non-debugging perl by using XS modules:
777 -Dr => use re 'debug'
779 -Dx => use O 'Debug'
780 Using a source-level debugger
782 If the debugging output of "-D" doesn't help you, it's time to step
783 through perl's execution with a source-level debugger.
784 • We'll use "gdb" for our examples here; the principles will apply to
786 any debugger (many vendors call their debugger "dbx"), but check the
787 manual of the one you're using.
788 To fire up the debugger, type
790 gdb ./perl
792 Or if you have a core dump:
794 gdb ./perl core
796 You'll want to do that in your Perl source tree so the debugger can
798 read the source code. You should see the copyright message, followed
799 by the prompt.
800 (gdb)
802 "help" will get you into the documentation, but here are the most
804 useful commands:
805 • run [args]
807 Run the program with the given arguments.
809 • break function_name
811 • break source.c:xxx
813 Tells the debugger that we'll want to pause execution when we reach
815 either the named function (but see "Internal Functions" in
816 perlguts!) or the given line in the named source file.
817 • step
819 Steps through the program a line at a time.
821 • next
823 Steps through the program a line at a time, without descending into
825 functions.
826 • continue
828 Run until the next breakpoint.
830 • finish
832 Run until the end of the current function, then stop again.
834 • 'enter'
836 Just pressing Enter will do the most recent operation again - it's a
838 blessing when stepping through miles of source code.
839 • ptype
841 Prints the C definition of the argument given.
843 (gdb) ptype PL_op
845 type = struct op {
846 OP *op_next;
847 OP *op_sibparent;
848 OP *(*op_ppaddr)(void);
849 PADOFFSET op_targ;
850 unsigned int op_type : 9;
851 unsigned int op_opt : 1;
852 unsigned int op_slabbed : 1;
853 unsigned int op_savefree : 1;
854 unsigned int op_static : 1;
855 unsigned int op_folded : 1;
856 unsigned int op_spare : 2;
857 U8 op_flags;
858 U8 op_private;
859 } *
860 • print
862 Execute the given C code and print its results. WARNING: Perl makes
864 heavy use of macros, and gdb does not necessarily support macros
865 (see later "gdb macro support"). You'll have to substitute them
866 yourself, or to invoke cpp on the source code files (see "The .i
867 Targets") So, for instance, you can't say
868 print SvPV_nolen(sv)
870 but you have to say
872 print Perl_sv_2pv_nolen(sv)
874 You may find it helpful to have a "macro dictionary", which you can
876 produce by saying "cpp -dM perl.c | sort". Even then, cpp won't
877 recursively apply those macros for you.
878 gdb macro support
880 Recent versions of gdb have fairly good macro support, but in order to
881 use it you'll need to compile perl with macro definitions included in
882 the debugging information. Using gcc version 3.1, this means
883 configuring with "-Doptimize=-g3". Other compilers might use a
884 different switch (if they support debugging macros at all).
885 Dumping Perl Data Structures
887 One way to get around this macro hell is to use the dumping functions
888 in dump.c; these work a little like an internal Devel::Peek, but they
889 also cover OPs and other structures that you can't get at from Perl.
890 Let's take an example. We'll use the "$a = $b + $c" we used before,
891 but give it a bit of context: "$b = "6XXXX"; $c = 2.3;". Where's a
892 good place to stop and poke around?
893 What about "pp_add", the function we examined earlier to implement the
895 "+" operator:
896 (gdb) break Perl_pp_add
898 Breakpoint 1 at 0x46249f: file pp_hot.c, line 309.
899 Notice we use "Perl_pp_add" and not "pp_add" - see "Internal Functions"
901 in perlguts. With the breakpoint in place, we can run our program:
902 (gdb) run -e '$b = "6XXXX"; $c = 2.3; $a = $b + $c'
904 Lots of junk will go past as gdb reads in the relevant source files and
906 libraries, and then:
907 Breakpoint 1, Perl_pp_add () at pp_hot.c:309
909 1396 dSP; dATARGET; bool useleft; SV *svl, *svr;
910 (gdb) step
911 311 dPOPTOPnnrl_ul;
912 (gdb)
913 We looked at this bit of code before, and we said that "dPOPTOPnnrl_ul"
915 arranges for two "NV"s to be placed into "left" and "right" - let's
916 slightly expand it:
917 #define dPOPTOPnnrl_ul NV right = POPn; \
919 SV *leftsv = TOPs; \
920 NV left = USE_LEFT(leftsv) ? SvNV(leftsv) : 0.0
921 "POPn" takes the SV from the top of the stack and obtains its NV either
923 directly (if "SvNOK" is set) or by calling the "sv_2nv" function.
924 "TOPs" takes the next SV from the top of the stack - yes, "POPn" uses
925 "TOPs" - but doesn't remove it. We then use "SvNV" to get the NV from
926 "leftsv" in the same way as before - yes, "POPn" uses "SvNV".
927 Since we don't have an NV for $b, we'll have to use "sv_2nv" to convert
929 it. If we step again, we'll find ourselves there:
930 (gdb) step
932 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1669
933 1669 if (!sv)
934 (gdb)
935 We can now use "Perl_sv_dump" to investigate the SV:
937 (gdb) print Perl_sv_dump(sv)
939 SV = PV(0xa057cc0) at 0xa0675d0
940 REFCNT = 1
941 FLAGS = (POK,pPOK)
942 PV = 0xa06a510 "6XXXX"\0
943 CUR = 5
944 LEN = 6
945 $1 = void
946 We know we're going to get 6 from this, so let's finish the subroutine:
948 (gdb) finish
950 Run till exit from #0 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1671
951 0x462669 in Perl_pp_add () at pp_hot.c:311
952 311 dPOPTOPnnrl_ul;
953 We can also dump out this op: the current op is always stored in
955 "PL_op", and we can dump it with "Perl_op_dump". This'll give us
956 similar output to CPAN module B::Debug.
957 (gdb) print Perl_op_dump(PL_op)
959 {
960 13 TYPE = add ===> 14
961 TARG = 1
962 FLAGS = (SCALAR,KIDS)
963 {
964 TYPE = null ===> (12)
965 (was rv2sv)
966 FLAGS = (SCALAR,KIDS)
967 {
968 11 TYPE = gvsv ===> 12
969 FLAGS = (SCALAR)
970 GV = main::b
971 }
972 }
973 # finish this later #
975 Using gdb to look at specific parts of a program
977 With the example above, you knew to look for "Perl_pp_add", but what if
978 there were multiple calls to it all over the place, or you didn't know
979 what the op was you were looking for?
980 One way to do this is to inject a rare call somewhere near what you're
982 looking for. For example, you could add "study" before your method:
983 study;
985 And in gdb do:
987 (gdb) break Perl_pp_study
989 And then step until you hit what you're looking for. This works well
991 in a loop if you want to only break at certain iterations:
992 for my $c (1..100) {
994 study if $c == 50;
995 }
996 Using gdb to look at what the parser/lexer are doing
998 If you want to see what perl is doing when parsing/lexing your code,
999 you can use "BEGIN {}":
1000 print "Before\n";
1002 BEGIN { study; }
1003 print "After\n";
1004 And in gdb:
1006 (gdb) break Perl_pp_study
1008 If you want to see what the parser/lexer is doing inside of "if" blocks
1010 and the like you need to be a little trickier:
1011 if ($a && $b && do { BEGIN { study } 1 } && $c) { ... }
1013
1015 Various tools exist for analysing C source code statically, as opposed
1016 to dynamically, that is, without executing the code. It is possible to
1017 detect resource leaks, undefined behaviour, type mismatches,
1018 portability problems, code paths that would cause illegal memory
1019 accesses, and other similar problems by just parsing the C code and
1020 looking at the resulting graph, what does it tell about the execution
1021 and data flows. As a matter of fact, this is exactly how C compilers
1022 know to give warnings about dubious code.
1023 lint
1025 The good old C code quality inspector, "lint", is available in several
1026 platforms, but please be aware that there are several different
1027 implementations of it by different vendors, which means that the flags
1028 are not identical across different platforms.
1029 There is a "lint" target in Makefile, but you may have to diddle with
1031 the flags (see above).
1032 Coverity
1034 Coverity (<http://www.coverity.com/>) is a product similar to lint and
1035 as a testbed for their product they periodically check several open
1036 source projects, and they give out accounts to open source developers
1037 to the defect databases.
1038 There is Coverity setup for the perl5 project:
1040 <https://scan.coverity.com/projects/perl5>
1041 HP-UX cadvise (Code Advisor)
1043 HP has a C/C++ static analyzer product for HP-UX caller Code Advisor.
1044 (Link not given here because the URL is horribly long and seems
1045 horribly unstable; use the search engine of your choice to find it.)
1046 The use of the "cadvise_cc" recipe with "Configure ...
1047 -Dcc=./cadvise_cc" (see cadvise "User Guide") is recommended; as is the
1048 use of "+wall".
1049 cpd (cut-and-paste detector)
1051 The cpd tool detects cut-and-paste coding. If one instance of the cut-
1052 and-pasted code changes, all the other spots should probably be
1053 changed, too. Therefore such code should probably be turned into a
1054 subroutine or a macro.
1055 cpd (<https://pmd.github.io/latest/pmd_userdocs_cpd.html>) is part of
1057 the pmd project (<https://pmd.github.io/>). pmd was originally written
1058 for static analysis of Java code, but later the cpd part of it was
1059 extended to parse also C and C++.
1060 Download the pmd-bin-X.Y.zip () from the SourceForge site, extract the
1062 pmd-X.Y.jar from it, and then run that on source code thusly:
1063 java -cp pmd-X.Y.jar net.sourceforge.pmd.cpd.CPD \
1065 --minimum-tokens 100 --files /some/where/src --language c > cpd.txt
1066 You may run into memory limits, in which case you should use the -Xmx
1068 option:
1069 java -Xmx512M ...
1071 gcc warnings
1073 Though much can be written about the inconsistency and coverage
1074 problems of gcc warnings (like "-Wall" not meaning "all the warnings",
1075 or some common portability problems not being covered by "-Wall", or
1076 "-ansi" and "-pedantic" both being a poorly defined collection of
1077 warnings, and so forth), gcc is still a useful tool in keeping our
1078 coding nose clean.
1079 The "-Wall" is by default on.
1081 It would be nice for "-pedantic") to be on always, but unfortunately it
1083 is not safe on all platforms - for example fatal conflicts with the
1084 system headers (Solaris being a prime example). If Configure
1085 "-Dgccansipedantic" is used, the "cflags" frontend selects "-pedantic"
1086 for the platforms where it is known to be safe.
1087 The following extra flags are added:
1089 • "-Wendif-labels"
1091 • "-Wextra"
1093 • "-Wc++-compat"
1095 • "-Wwrite-strings"
1097 • "-Werror=pointer-arith"
1099 • "-Werror=vla"
1101 The following flags would be nice to have but they would first need
1103 their own Augean stablemaster:
1104 • "-Wshadow"
1106 • "-Wstrict-prototypes"
1108 The "-Wtraditional" is another example of the annoying tendency of gcc
1110 to bundle a lot of warnings under one switch (it would be impossible to
1111 deploy in practice because it would complain a lot) but it does contain
1112 some warnings that would be beneficial to have available on their own,
1113 such as the warning about string constants inside macros containing the
1114 macro arguments: this behaved differently pre-ANSI than it does in
1115 ANSI, and some C compilers are still in transition, AIX being an
1116 example.
1117 Warnings of other C compilers
1119 Other C compilers (yes, there are other C compilers than gcc) often
1120 have their "strict ANSI" or "strict ANSI with some portability
1121 extensions" modes on, like for example the Sun Workshop has its "-Xa"
1122 mode on (though implicitly), or the DEC (these days, HP...) has its
1123 "-std1" mode on.
1124
1126 NOTE 1: Running under older memory debuggers such as Purify, valgrind
1127 or Third Degree greatly slows down the execution: seconds become
1128 minutes, minutes become hours. For example as of Perl 5.8.1, the
1129 ext/Encode/t/Unicode.t takes extraordinarily long to complete under
1130 e.g. Purify, Third Degree, and valgrind. Under valgrind it takes more
1131 than six hours, even on a snappy computer. The said test must be doing
1132 something that is quite unfriendly for memory debuggers. If you don't
1133 feel like waiting, that you can simply kill away the perl process.
1134 Roughly valgrind slows down execution by factor 10, AddressSanitizer by
1135 factor 2.
1136 NOTE 2: To minimize the number of memory leak false alarms (see
1138 "PERL_DESTRUCT_LEVEL" for more information), you have to set the
1139 environment variable PERL_DESTRUCT_LEVEL to 2. For example, like this:
1140 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib ...
1142 NOTE 3: There are known memory leaks when there are compile-time errors
1144 within eval or require, seeing "S_doeval" in the call stack is a good
1145 sign of these. Fixing these leaks is non-trivial, unfortunately, but
1146 they must be fixed eventually.
1147 NOTE 4: DynaLoader will not clean up after itself completely unless
1149 Perl is built with the Configure option
1150 "-Accflags=-DDL_UNLOAD_ALL_AT_EXIT".
1151 valgrind
1153 The valgrind tool can be used to find out both memory leaks and illegal
1154 heap memory accesses. As of version 3.3.0, Valgrind only supports
1155 Linux on x86, x86-64 and PowerPC and Darwin (OS X) on x86 and x86-64.
1156 The special "test.valgrind" target can be used to run the tests under
1157 valgrind. Found errors and memory leaks are logged in files named
1158 testfile.valgrind and by default output is displayed inline.
1159 Example usage:
1161 make test.valgrind
1163 Since valgrind adds significant overhead, tests will take much longer
1165 to run. The valgrind tests support being run in parallel to help with
1166 this:
1167 TEST_JOBS=9 make test.valgrind
1169 Note that the above two invocations will be very verbose as reachable
1171 memory and leak-checking is enabled by default. If you want to just
1172 see pure errors, try:
1173 VG_OPTS='-q --leak-check=no --show-reachable=no' TEST_JOBS=9 \
1175 make test.valgrind
1176 Valgrind also provides a cachegrind tool, invoked on perl as:
1178 VG_OPTS=--tool=cachegrind make test.valgrind
1180 As system libraries (most notably glibc) are also triggering errors,
1182 valgrind allows to suppress such errors using suppression files. The
1183 default suppression file that comes with valgrind already catches a lot
1184 of them. Some additional suppressions are defined in t/perl.supp.
1185 To get valgrind and for more information see
1187 http://valgrind.org/
1189 AddressSanitizer
1191 AddressSanitizer ("ASan") consists of a compiler instrumentation module
1192 and a run-time "malloc" library. ASan is available for a variety of
1193 architectures, operating systems, and compilers (see project link
1194 below). It checks for unsafe memory usage, such as use after free and
1195 buffer overflow conditions, and is fast enough that you can easily
1196 compile your debugging or optimized perl with it. Modern versions of
1197 ASan check for memory leaks by default on most platforms, otherwise
1198 (e.g. x86_64 OS X) this feature can be enabled via
1199 "ASAN_OPTIONS=detect_leaks=1".
1200 To build perl with AddressSanitizer, your Configure invocation should
1202 look like:
1203 sh Configure -des -Dcc=clang \
1205 -Accflags=-fsanitize=address -Aldflags=-fsanitize=address \
1206 -Alddlflags=-shared\ -fsanitize=address \
1207 -fsanitize-blacklist=`pwd`/asan_ignore
1208 where these arguments mean:
1210 • -Dcc=clang
1212 This should be replaced by the full path to your clang executable
1214 if it is not in your path.
1215 • -Accflags=-fsanitize=address
1217 Compile perl and extensions sources with AddressSanitizer.
1219 • -Aldflags=-fsanitize=address
1221 Link the perl executable with AddressSanitizer.
1223 • -Alddlflags=-shared\ -fsanitize=address
1225 Link dynamic extensions with AddressSanitizer. You must manually
1227 specify "-shared" because using "-Alddlflags=-shared" will prevent
1228 Configure from setting a default value for "lddlflags", which
1229 usually contains "-shared" (at least on Linux).
1230 • -fsanitize-blacklist=`pwd`/asan_ignore
1232 AddressSanitizer will ignore functions listed in the "asan_ignore"
1234 file. (This file should contain a short explanation of why each of
1235 the functions is listed.)
1236 See also <https://github.com/google/sanitizers/wiki/AddressSanitizer>.
1238
1240 Depending on your platform there are various ways of profiling Perl.
1241 There are two commonly used techniques of profiling executables:
1243 statistical time-sampling and basic-block counting.
1244 The first method takes periodically samples of the CPU program counter,
1246 and since the program counter can be correlated with the code generated
1247 for functions, we get a statistical view of in which functions the
1248 program is spending its time. The caveats are that very small/fast
1249 functions have lower probability of showing up in the profile, and that
1250 periodically interrupting the program (this is usually done rather
1251 frequently, in the scale of milliseconds) imposes an additional
1252 overhead that may skew the results. The first problem can be
1253 alleviated by running the code for longer (in general this is a good
1254 idea for profiling), the second problem is usually kept in guard by the
1255 profiling tools themselves.
1256 The second method divides up the generated code into basic blocks.
1258 Basic blocks are sections of code that are entered only in the
1259 beginning and exited only at the end. For example, a conditional jump
1260 starts a basic block. Basic block profiling usually works by
1261 instrumenting the code by adding enter basic block #nnnn book-keeping
1262 code to the generated code. During the execution of the code the basic
1263 block counters are then updated appropriately. The caveat is that the
1264 added extra code can skew the results: again, the profiling tools
1265 usually try to factor their own effects out of the results.
1266 Gprof Profiling
1268 gprof is a profiling tool available in many Unix platforms which uses
1269 statistical time-sampling. You can build a profiled version of perl by
1270 compiling using gcc with the flag "-pg". Either edit config.sh or re-
1271 run Configure. Running the profiled version of Perl will create an
1272 output file called gmon.out which contains the profiling data collected
1273 during the execution.
1274 quick hint:
1276 $ sh Configure -des -Dusedevel -Accflags='-pg' \
1278 -Aldflags='-pg' -Alddlflags='-pg -shared' \
1279 && make perl
1280 $ ./perl ... # creates gmon.out in current directory
1281 $ gprof ./perl > out
1282 $ less out
1283 (you probably need to add "-shared" to the <-Alddlflags> line until RT
1285 #118199 is resolved)
1286 The gprof tool can then display the collected data in various ways.
1288 Usually gprof understands the following options:
1289 • -a
1291 Suppress statically defined functions from the profile.
1293 • -b
1295 Suppress the verbose descriptions in the profile.
1297 • -e routine
1299 Exclude the given routine and its descendants from the profile.
1301 • -f routine
1303 Display only the given routine and its descendants in the profile.
1305 • -s
1307 Generate a summary file called gmon.sum which then may be given to
1309 subsequent gprof runs to accumulate data over several runs.
1310 • -z
1312 Display routines that have zero usage.
1314 For more detailed explanation of the available commands and output
1316 formats, see your own local documentation of gprof.
1317 GCC gcov Profiling
1319 basic block profiling is officially available in gcc 3.0 and later.
1320 You can build a profiled version of perl by compiling using gcc with
1321 the flags "-fprofile-arcs -ftest-coverage". Either edit config.sh or
1322 re-run Configure.
1323 quick hint:
1325 $ sh Configure -des -Dusedevel -Doptimize='-g' \
1327 -Accflags='-fprofile-arcs -ftest-coverage' \
1328 -Aldflags='-fprofile-arcs -ftest-coverage' \
1329 -Alddlflags='-fprofile-arcs -ftest-coverage -shared' \
1330 && make perl
1331 $ rm -f regexec.c.gcov regexec.gcda
1332 $ ./perl ...
1333 $ gcov regexec.c
1334 $ less regexec.c.gcov
1335 (you probably need to add "-shared" to the <-Alddlflags> line until RT
1337 #118199 is resolved)
1338 Running the profiled version of Perl will cause profile output to be
1340 generated. For each source file an accompanying .gcda file will be
1341 created.
1342 To display the results you use the gcov utility (which should be
1344 installed if you have gcc 3.0 or newer installed). gcov is run on
1345 source code files, like this
1346 gcov sv.c
1348 which will cause sv.c.gcov to be created. The .gcov files contain the
1350 source code annotated with relative frequencies of execution indicated
1351 by "#" markers. If you want to generate .gcov files for all profiled
1352 object files, you can run something like this:
1353 for file in `find . -name \*.gcno`
1355 do sh -c "cd `dirname $file` && gcov `basename $file .gcno`"
1356 done
1357 Useful options of gcov include "-b" which will summarise the basic
1359 block, branch, and function call coverage, and "-c" which instead of
1360 relative frequencies will use the actual counts. For more information
1361 on the use of gcov and basic block profiling with gcc, see the latest
1362 GNU CC manual. As of gcc 4.8, this is at
1363 <http://gcc.gnu.org/onlinedocs/gcc/Gcov-Intro.html#Gcov-Intro>
1364 callgrind profiling
1366 callgrind is a valgrind tool for profiling source code. Paired with
1367 kcachegrind (a Qt based UI), it gives you an overview of where code is
1368 taking up time, as well as the ability to examine callers, call trees,
1369 and more. One of its benefits is you can use it on perl and XS modules
1370 that have not been compiled with debugging symbols.
1371 If perl is compiled with debugging symbols ("-g"), you can view the
1373 annotated source and click around, much like Devel::NYTProf's HTML
1374 output.
1375 For basic usage:
1377 valgrind --tool=callgrind ./perl ...
1379 By default it will write output to callgrind.out.PID, but you can
1381 change that with "--callgrind-out-file=..."
1382 To view the data, do:
1384 kcachegrind callgrind.out.PID
1386 If you'd prefer to view the data in a terminal, you can use
1388 callgrind_annotate. In it's basic form:
1389 callgrind_annotate callgrind.out.PID | less
1391 Some useful options are:
1393 • --threshold
1395 Percentage of counts (of primary sort event) we are interested in.
1397 The default is 99%, 100% might show things that seem to be missing.
1398 • --auto
1400 Annotate all source files containing functions that helped reach
1402 the event count threshold.
1403
1405 PERL_DESTRUCT_LEVEL
1406 If you want to run any of the tests yourself manually using e.g.
1407 valgrind, please note that by default perl does not explicitly cleanup
1408 all the memory it has allocated (such as global memory arenas) but
1409 instead lets the exit() of the whole program "take care" of such
1410 allocations, also known as "global destruction of objects".
1411 There is a way to tell perl to do complete cleanup: set the environment
1413 variable PERL_DESTRUCT_LEVEL to a non-zero value. The t/TEST wrapper
1414 does set this to 2, and this is what you need to do too, if you don't
1415 want to see the "global leaks": For example, for running under valgrind
1416 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib t/foo/bar.t
1418 (Note: the mod_perl apache module uses also this environment variable
1420 for its own purposes and extended its semantics. Refer to the mod_perl
1421 documentation for more information. Also, spawned threads do the
1422 equivalent of setting this variable to the value 1.)
1423 If, at the end of a run you get the message N scalars leaked, you can
1425 recompile with "-DDEBUG_LEAKING_SCALARS", ("Configure
1426 -Accflags=-DDEBUG_LEAKING_SCALARS"), which will cause the addresses of
1427 all those leaked SVs to be dumped along with details as to where each
1428 SV was originally allocated. This information is also displayed by
1429 Devel::Peek. Note that the extra details recorded with each SV
1430 increases memory usage, so it shouldn't be used in production
1431 environments. It also converts "new_SV()" from a macro into a real
1432 function, so you can use your favourite debugger to discover where
1433 those pesky SVs were allocated.
1434 If you see that you're leaking memory at runtime, but neither valgrind
1436 nor "-DDEBUG_LEAKING_SCALARS" will find anything, you're probably
1437 leaking SVs that are still reachable and will be properly cleaned up
1438 during destruction of the interpreter. In such cases, using the "-Dm"
1439 switch can point you to the source of the leak. If the executable was
1440 built with "-DDEBUG_LEAKING_SCALARS", "-Dm" will output SV allocations
1441 in addition to memory allocations. Each SV allocation has a distinct
1442 serial number that will be written on creation and destruction of the
1443 SV. So if you're executing the leaking code in a loop, you need to
1444 look for SVs that are created, but never destroyed between each cycle.
1445 If such an SV is found, set a conditional breakpoint within "new_SV()"
1446 and make it break only when "PL_sv_serial" is equal to the serial
1447 number of the leaking SV. Then you will catch the interpreter in
1448 exactly the state where the leaking SV is allocated, which is
1449 sufficient in many cases to find the source of the leak.
1450 As "-Dm" is using the PerlIO layer for output, it will by itself
1452 allocate quite a bunch of SVs, which are hidden to avoid recursion.
1453 You can bypass the PerlIO layer if you use the SV logging provided by
1454 "-DPERL_MEM_LOG" instead.
1455 PERL_MEM_LOG
1457 If compiled with "-DPERL_MEM_LOG" ("-Accflags=-DPERL_MEM_LOG"), both
1458 memory and SV allocations go through logging functions, which is handy
1459 for breakpoint setting.
1460 Unless "-DPERL_MEM_LOG_NOIMPL" ("-Accflags=-DPERL_MEM_LOG_NOIMPL") is
1462 also compiled, the logging functions read $ENV{PERL_MEM_LOG} to
1463 determine whether to log the event, and if so how:
1464 $ENV{PERL_MEM_LOG} =~ /m/ Log all memory ops
1466 $ENV{PERL_MEM_LOG} =~ /s/ Log all SV ops
1467 $ENV{PERL_MEM_LOG} =~ /t/ include timestamp in Log
1468 $ENV{PERL_MEM_LOG} =~ /^(\d+)/ write to FD given (default is 2)
1469 Memory logging is somewhat similar to "-Dm" but is independent of
1471 "-DDEBUGGING", and at a higher level; all uses of Newx(), Renew(), and
1472 Safefree() are logged with the caller's source code file and line
1473 number (and C function name, if supported by the C compiler). In
1474 contrast, "-Dm" is directly at the point of "malloc()". SV logging is
1475 similar.
1476 Since the logging doesn't use PerlIO, all SV allocations are logged and
1478 no extra SV allocations are introduced by enabling the logging. If
1479 compiled with "-DDEBUG_LEAKING_SCALARS", the serial number for each SV
1480 allocation is also logged.
1481 DDD over gdb
1483 Those debugging perl with the DDD frontend over gdb may find the
1484 following useful:
1485 You can extend the data conversion shortcuts menu, so for example you
1487 can display an SV's IV value with one click, without doing any typing.
1488 To do that simply edit ~/.ddd/init file and add after:
1489 ! Display shortcuts.
1491 Ddd*gdbDisplayShortcuts: \
1492 /t () // Convert to Bin\n\
1493 /d () // Convert to Dec\n\
1494 /x () // Convert to Hex\n\
1495 /o () // Convert to Oct(\n\
1496 the following two lines:
1498 ((XPV*) (())->sv_any )->xpv_pv // 2pvx\n\
1500 ((XPVIV*) (())->sv_any )->xiv_iv // 2ivx
1501 so now you can do ivx and pvx lookups or you can plug there the sv_peek
1503 "conversion":
1504 Perl_sv_peek(my_perl, (SV*)()) // sv_peek
1506 (The my_perl is for threaded builds.) Just remember that every line,
1508 but the last one, should end with \n\
1509 Alternatively edit the init file interactively via: 3rd mouse button ->
1511 New Display -> Edit Menu
1512 Note: you can define up to 20 conversion shortcuts in the gdb section.
1514 C backtrace
1516 On some platforms Perl supports retrieving the C level backtrace
1517 (similar to what symbolic debuggers like gdb do).
1518 The backtrace returns the stack trace of the C call frames, with the
1520 symbol names (function names), the object names (like "perl"), and if
1521 it can, also the source code locations (file:line).
1522 The supported platforms are Linux, and OS X (some *BSD might work at
1524 least partly, but they have not yet been tested).
1525 This feature hasn't been tested with multiple threads, but it will only
1527 show the backtrace of the thread doing the backtracing.
1528 The feature needs to be enabled with "Configure -Dusecbacktrace".
1530 The "-Dusecbacktrace" also enables keeping the debug information when
1532 compiling/linking (often: "-g"). Many compilers/linkers do support
1533 having both optimization and keeping the debug information. The debug
1534 information is needed for the symbol names and the source locations.
1535 Static functions might not be visible for the backtrace.
1537 Source code locations, even if available, can often be missing or
1539 misleading if the compiler has e.g. inlined code. Optimizer can make
1540 matching the source code and the object code quite challenging.
1541 Linux
1543 You must have the BFD (-lbfd) library installed, otherwise "perl"
1544 will fail to link. The BFD is usually distributed as part of the
1545 GNU binutils.
1546 Summary: "Configure ... -Dusecbacktrace" and you need "-lbfd".
1548 OS X
1550 The source code locations are supported only if you have the
1551 Developer Tools installed. (BFD is not needed.)
1552 Summary: "Configure ... -Dusecbacktrace" and installing the
1554 Developer Tools would be good.
1555 Optionally, for trying out the feature, you may want to enable
1557 automatic dumping of the backtrace just before a warning or croak (die)
1558 message is emitted, by adding "-Accflags=-DUSE_C_BACKTRACE_ON_ERROR"
1559 for Configure.
1560 Unless the above additional feature is enabled, nothing about the
1562 backtrace functionality is visible, except for the Perl/XS level.
1563 Furthermore, even if you have enabled this feature to be compiled, you
1565 need to enable it in runtime with an environment variable:
1566 "PERL_C_BACKTRACE_ON_ERROR=10". It must be an integer higher than
1567 zero, telling the desired frame count.
1568 Retrieving the backtrace from Perl level (using for example an XS
1570 extension) would be much less exciting than one would hope: normally
1571 you would see "runops", "entersub", and not much else. This API is
1572 intended to be called from within the Perl implementation, not from
1573 Perl level execution.
1574 The C API for the backtrace is as follows:
1576 get_c_backtrace
1578 free_c_backtrace
1579 get_c_backtrace_dump
1580 dump_c_backtrace
1581 Poison
1583 If you see in a debugger a memory area mysteriously full of 0xABABABAB
1584 or 0xEFEFEFEF, you may be seeing the effect of the Poison() macros, see
1585 perlclib.
1586 Read-only optrees
1588 Under ithreads the optree is read only. If you want to enforce this,
1589 to check for write accesses from buggy code, compile with
1590 "-Accflags=-DPERL_DEBUG_READONLY_OPS" to enable code that allocates op
1591 memory via "mmap", and sets it read-only when it is attached to a
1592 subroutine. Any write access to an op results in a "SIGBUS" and abort.
1593 This code is intended for development only, and may not be portable
1595 even to all Unix variants. Also, it is an 80% solution, in that it
1596 isn't able to make all ops read only. Specifically it does not apply
1597 to op slabs belonging to "BEGIN" blocks.
1598 However, as an 80% solution it is still effective, as it has caught
1600 bugs in the past.
1601 When is a bool not a bool?
1603 There wasn't necessarily a standard "bool" type on compilers prior to
1604 C99, and so some workarounds were created. The "TRUE" and "FALSE"
1605 macros are still available as alternatives for "true" and "false". And
1606 the "cBOOL" macro was created to correctly cast to a true/false value
1607 in all circumstances, but should no longer be necessary. Using
1608 "(bool)" expr> should now always work.
1609 There are no plans to remove any of "TRUE", "FALSE", nor "cBOOL".
1611 Finding unsafe truncations
1613 You may wish to run "Configure" with something like
1614 -Accflags='-Wconversion -Wno-sign-conversion -Wno-shorten-64-to-32'
1616 or your compiler's equivalent to make it easier to spot any unsafe
1618 truncations that show up.
1619 The .i Targets
1621 You can expand the macros in a foo.c file by saying
1622 make foo.i
1624 which will expand the macros using cpp. Don't be scared by the
1626 results.
1627
1629 This document was originally written by Nathan Torkington, and is
1630 maintained by the perl5-porters mailing list.
1631perl v5.36.0 2022-08-30 PERLHACKTIPS(1)