1PERLHACKTIPS(1) Perl Programmers Reference Guide PERLHACKTIPS(1)
2
3
4
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
13 If you haven't read perlhack and perlhacktut yet, you might want to do
14 that first.
15
17 Perl source plays by ANSI C89 rules: no C99 (or C++) extensions. In
18 some cases we have to take pre-ANSI requirements into consideration.
19 You don't care about some particular platform having broken Perl? I
20 hear there is still a strong demand for J2EE programmers.
21
22 Perl environment problems
23 · Not compiling with threading
24
25 Compiling with threading (-Duseithreads) completely rewrites the
26 function prototypes of Perl. You better try your changes with that.
27 Related to this is the difference between "Perl_-less" and
28 "Perl_-ly" APIs, for example:
29
30 Perl_sv_setiv(aTHX_ ...);
31 sv_setiv(...);
32
33 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 to
36 do a dTHX (or a dVAR) as the first thing in the function.
37
38 See "How multiple interpreters and concurrency are supported" in
39 perlguts for further discussion about context.
40
41 · Not compiling with -DDEBUGGING
42
43 The DEBUGGING define exposes more code to the compiler, therefore
44 more ways for things to go wrong. You should try it.
45
46 · Introducing (non-read-only) globals
47
48 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
54 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
59 If you want to have static strings, make them constant:
60
61 static const char etc[] = "...";
62
63 If you want to have arrays of constant strings, note carefully the
64 right combination of "const"s:
65
66 static const char * const yippee[] =
67 {"hi", "ho", "silver"};
68
69 There is a way to completely hide any modifiable globals (they are
70 all moved to heap), the compilation setting
71 "-DPERL_GLOBAL_STRUCT_PRIVATE". It is not normally used, but can be
72 used for testing, read more about it in "Background and
73 PERL_IMPLICIT_CONTEXT" in perlguts.
74
75 · Not exporting your new function
76
77 Some platforms (Win32, AIX, VMS, OS/2, to name a few) require any
78 function that is part of the public API (the shared Perl library)
79 to be explicitly marked as exported. See the discussion about
80 embed.pl in perlguts.
81
82 · Exporting your new function
83
84 The new shiny result of either genuine new functionality or your
85 arduous refactoring is now ready and correctly exported. So what
86 could possibly go wrong?
87
88 Maybe simply that your function did not need to be exported in the
89 first place. Perl has a long and not so glorious history of
90 exporting functions that it should not have.
91
92 If the function is used only inside one source code file, make it
93 static. See the discussion about embed.pl in perlguts.
94
95 If the function is used across several files, but intended only for
96 Perl's internal use (and this should be the common case), do not
97 export it to the public API. See the discussion about embed.pl in
98 perlguts.
99
100 Portability problems
101 The following are common causes of compilation and/or execution
102 failures, not common to Perl as such. The C FAQ is good bedtime
103 reading. Please test your changes with as many C compilers and
104 platforms as possible; we will, anyway, and it's nice to save oneself
105 from public embarrassment.
106
107 If using gcc, you can add the "-std=c89" option which will hopefully
108 catch most of these unportabilities. (However it might also catch
109 incompatibilities in your system's header files.)
110
111 Use the Configure "-Dgccansipedantic" flag to enable the gcc "-ansi
112 -pedantic" flags which enforce stricter ANSI rules.
113
114 If using the "gcc -Wall" note that not all the possible warnings (like
115 "-Wunitialized") are given unless you also compile with "-O".
116
117 Note that if using gcc, starting from Perl 5.9.5 the Perl core source
118 code files (the ones at the top level of the source code distribution,
119 but not e.g. the extensions under ext/) are automatically compiled with
120 as many as possible of the "-std=c89", "-ansi", "-pedantic", and a
121 selection of "-W" flags (see cflags.SH).
122
123 Also study perlport carefully to avoid any bad assumptions about the
124 operating system, filesystems, and so forth.
125
126 You may once in a while try a "make microperl" to see whether we can
127 still compile Perl with just the bare minimum of interfaces. (See
128 README.micro.)
129
130 Do not assume an operating system indicates a certain compiler.
131
132 · Casting pointers to integers or casting integers to pointers
133
134 void castaway(U8* p)
135 {
136 IV i = p;
137
138 or
139
140 void castaway(U8* p)
141 {
142 IV i = (IV)p;
143
144 Both are bad, and broken, and unportable. Use the PTR2IV() macro
145 that does it right. (Likewise, there are PTR2UV(), PTR2NV(),
146 INT2PTR(), and NUM2PTR().)
147
148 · Casting between data function pointers and data pointers
149
150 Technically speaking casting between function pointers and data
151 pointers is unportable and undefined, but practically speaking it
152 seems to work, but you should use the FPTR2DPTR() and DPTR2FPTR()
153 macros. Sometimes you can also play games with unions.
154
155 · Assuming sizeof(int) == sizeof(long)
156
157 There are platforms where longs are 64 bits, and platforms where
158 ints are 64 bits, and while we are out to shock you, even platforms
159 where shorts are 64 bits. This is all legal according to the C
160 standard. (In other words, "long long" is not a portable way to
161 specify 64 bits, and "long long" is not even guaranteed to be any
162 wider than "long".)
163
164 Instead, use the definitions IV, UV, IVSIZE, I32SIZE, and so forth.
165 Avoid things like I32 because they are not guaranteed to be exactly
166 32 bits, they are at least 32 bits, nor are they guaranteed to be
167 int or long. If you really explicitly need 64-bit variables, use
168 I64 and U64, but only if guarded by HAS_QUAD.
169
170 · Assuming one can dereference any type of pointer for any type of
171 data
172
173 char *p = ...;
174 long pony = *p; /* BAD */
175
176 Many platforms, quite rightly so, will give you a core dump instead
177 of a pony if the p happens not to be correctly aligned.
178
179 · Lvalue casts
180
181 (int)*p = ...; /* BAD */
182
183 Simply not portable. Get your lvalue to be of the right type, or
184 maybe use temporary variables, or dirty tricks with unions.
185
186 · Assume anything about structs (especially the ones you don't
187 control, like the ones coming from the system headers)
188
189 · That a certain field exists in a struct
190
191 · That no other fields exist besides the ones you know of
192
193 · That a field is of certain signedness, sizeof, or type
194
195 · That the fields are in a certain order
196
197 · While C guarantees the ordering specified in the
198 struct definition, between different platforms the
199 definitions might differ
200
201 · That the sizeof(struct) or the alignments are the same
202 everywhere
203
204 · There might be padding bytes between the fields to
205 align the fields - the bytes can be anything
206
207 · Structs are required to be aligned to the maximum
208 alignment required by the fields - which for native
209 types is for usually equivalent to sizeof() of the
210 field
211
212 · Assuming the character set is ASCIIish
213
214 Perl can compile and run under EBCDIC platforms. See perlebcdic.
215 This is transparent for the most part, but because the character
216 sets differ, you shouldn't use numeric (decimal, octal, nor hex)
217 constants to refer to characters. You can safely say 'A', but not
218 0x41. You can safely say '\n', but not \012. If a character doesn't
219 have a trivial input form, you can create a #define for it in both
220 "utfebcdic.h" and "utf8.h", so that it resolves to different values
221 depending on the character set being used. (There are three
222 different EBCDIC character sets defined in "utfebcdic.h", so it
223 might be best to insert the #define three times in that file.)
224
225 Also, the range 'A' - 'Z' in ASCII is an unbroken sequence of 26
226 upper case alphabetic characters. That is not true in EBCDIC. Nor
227 for 'a' to 'z'. But '0' - '9' is an unbroken range in both systems.
228 Don't assume anything about other ranges.
229
230 Many of the comments in the existing code ignore the possibility of
231 EBCDIC, and may be wrong therefore, even if the code works. This is
232 actually a tribute to the successful transparent insertion of being
233 able to handle EBCDIC without having to change pre-existing code.
234
235 UTF-8 and UTF-EBCDIC are two different encodings used to represent
236 Unicode code points as sequences of bytes. Macros with the same
237 names (but different definitions) in "utf8.h" and "utfebcdic.h" are
238 used to allow the calling code to think that there is only one such
239 encoding. This is almost always referred to as "utf8", but it
240 means the EBCDIC version as well. Again, comments in the code may
241 well be wrong even if the code itself is right. For example, the
242 concept of "invariant characters" differs between ASCII and EBCDIC.
243 On ASCII platforms, only characters that do not have the high-order
244 bit set (i.e. whose ordinals are strict ASCII, 0 - 127) are
245 invariant, and the documentation and comments in the code may
246 assume that, often referring to something like, say, "hibit". The
247 situation differs and is not so simple on EBCDIC machines, but as
248 long as the code itself uses the "NATIVE_IS_INVARIANT()" macro
249 appropriately, it works, even if the comments are wrong.
250
251 · Assuming the character set is just ASCII
252
253 ASCII is a 7 bit encoding, but bytes have 8 bits in them. The 128
254 extra characters have different meanings depending on the locale.
255 Absent a locale, currently these extra characters are generally
256 considered to be unassigned, and this has presented some problems.
257 This is being changed starting in 5.12 so that these characters
258 will be considered to be Latin-1 (ISO-8859-1).
259
260 · Mixing #define and #ifdef
261
262 #define BURGLE(x) ... \
263 #ifdef BURGLE_OLD_STYLE /* BAD */
264 ... do it the old way ... \
265 #else
266 ... do it the new way ... \
267 #endif
268
269 You cannot portably "stack" cpp directives. For example in the
270 above you need two separate BURGLE() #defines, one for each #ifdef
271 branch.
272
273 · Adding non-comment stuff after #endif or #else
274
275 #ifdef SNOSH
276 ...
277 #else !SNOSH /* BAD */
278 ...
279 #endif SNOSH /* BAD */
280
281 The #endif and #else cannot portably have anything non-comment
282 after them. If you want to document what is going (which is a good
283 idea especially if the branches are long), use (C) comments:
284
285 #ifdef SNOSH
286 ...
287 #else /* !SNOSH */
288 ...
289 #endif /* SNOSH */
290
291 The gcc option "-Wendif-labels" warns about the bad variant (by
292 default on starting from Perl 5.9.4).
293
294 · Having a comma after the last element of an enum list
295
296 enum color {
297 CERULEAN,
298 CHARTREUSE,
299 CINNABAR, /* BAD */
300 };
301
302 is not portable. Leave out the last comma.
303
304 Also note that whether enums are implicitly morphable to ints
305 varies between compilers, you might need to (int).
306
307 · Using //-comments
308
309 // This function bamfoodles the zorklator. /* BAD */
310
311 That is C99 or C++. Perl is C89. Using the //-comments is silently
312 allowed by many C compilers but cranking up the ANSI C89 strictness
313 (which we like to do) causes the compilation to fail.
314
315 · Mixing declarations and code
316
317 void zorklator()
318 {
319 int n = 3;
320 set_zorkmids(n); /* BAD */
321 int q = 4;
322
323 That is C99 or C++. Some C compilers allow that, but you shouldn't.
324
325 The gcc option "-Wdeclaration-after-statements" scans for such
326 problems (by default on starting from Perl 5.9.4).
327
328 · Introducing variables inside for()
329
330 for(int i = ...; ...; ...) { /* BAD */
331
332 That is C99 or C++. While it would indeed be awfully nice to have
333 that also in C89, to limit the scope of the loop variable, alas, we
334 cannot.
335
336 · Mixing signed char pointers with unsigned char pointers
337
338 int foo(char *s) { ... }
339 ...
340 unsigned char *t = ...; /* Or U8* t = ... */
341 foo(t); /* BAD */
342
343 While this is legal practice, it is certainly dubious, and
344 downright fatal in at least one platform: for example VMS cc
345 considers this a fatal error. One cause for people often making
346 this mistake is that a "naked char" and therefore dereferencing a
347 "naked char pointer" have an undefined signedness: it depends on
348 the compiler and the flags of the compiler and the underlying
349 platform whether the result is signed or unsigned. For this very
350 same reason using a 'char' as an array index is bad.
351
352 · Macros that have string constants and their arguments as substrings
353 of the string constants
354
355 #define FOO(n) printf("number = %d\n", n) /* BAD */
356 FOO(10);
357
358 Pre-ANSI semantics for that was equivalent to
359
360 printf("10umber = %d\10");
361
362 which is probably not what you were expecting. Unfortunately at
363 least one reasonably common and modern C compiler does "real
364 backward compatibility" here, in AIX that is what still happens
365 even though the rest of the AIX compiler is very happily C89.
366
367 · Using printf formats for non-basic C types
368
369 IV i = ...;
370 printf("i = %d\n", i); /* BAD */
371
372 While this might by accident work in some platform (where IV
373 happens to be an "int"), in general it cannot. IV might be
374 something larger. Even worse the situation is with more specific
375 types (defined by Perl's configuration step in config.h):
376
377 Uid_t who = ...;
378 printf("who = %d\n", who); /* BAD */
379
380 The problem here is that Uid_t might be not only not "int"-wide but
381 it might also be unsigned, in which case large uids would be
382 printed as negative values.
383
384 There is no simple solution to this because of printf()'s limited
385 intelligence, but for many types the right format is available as
386 with either 'f' or '_f' suffix, for example:
387
388 IVdf /* IV in decimal */
389 UVxf /* UV is hexadecimal */
390
391 printf("i = %"IVdf"\n", i); /* The IVdf is a string constant. */
392
393 Uid_t_f /* Uid_t in decimal */
394
395 printf("who = %"Uid_t_f"\n", who);
396
397 Or you can try casting to a "wide enough" type:
398
399 printf("i = %"IVdf"\n", (IV)something_very_small_and_signed);
400
401 Also remember that the %p format really does require a void
402 pointer:
403
404 U8* p = ...;
405 printf("p = %p\n", (void*)p);
406
407 The gcc option "-Wformat" scans for such problems.
408
409 · Blindly using variadic macros
410
411 gcc has had them for a while with its own syntax, and C99 brought
412 them with a standardized syntax. Don't use the former, and use the
413 latter only if the HAS_C99_VARIADIC_MACROS is defined.
414
415 · Blindly passing va_list
416
417 Not all platforms support passing va_list to further varargs
418 (stdarg) functions. The right thing to do is to copy the va_list
419 using the Perl_va_copy() if the NEED_VA_COPY is defined.
420
421 · Using gcc statement expressions
422
423 val = ({...;...;...}); /* BAD */
424
425 While a nice extension, it's not portable. The Perl code does
426 admittedly use them if available to gain some extra speed
427 (essentially as a funky form of inlining), but you shouldn't.
428
429 · Binding together several statements in a macro
430
431 Use the macros STMT_START and STMT_END.
432
433 STMT_START {
434 ...
435 } STMT_END
436
437 · Testing for operating systems or versions when should be testing
438 for features
439
440 #ifdef __FOONIX__ /* BAD */
441 foo = quux();
442 #endif
443
444 Unless you know with 100% certainty that quux() is only ever
445 available for the "Foonix" operating system and that is available
446 and correctly working for all past, present, and future versions of
447 "Foonix", the above is very wrong. This is more correct (though
448 still not perfect, because the below is a compile-time check):
449
450 #ifdef HAS_QUUX
451 foo = quux();
452 #endif
453
454 How does the HAS_QUUX become defined where it needs to be? Well,
455 if Foonix happens to be Unixy enough to be able to run the
456 Configure script, and Configure has been taught about detecting and
457 testing quux(), the HAS_QUUX will be correctly defined. In other
458 platforms, the corresponding configuration step will hopefully do
459 the same.
460
461 In a pinch, if you cannot wait for Configure to be educated, or if
462 you have a good hunch of where quux() might be available, you can
463 temporarily try the following:
464
465 #if (defined(__FOONIX__) || defined(__BARNIX__))
466 # define HAS_QUUX
467 #endif
468
469 ...
470
471 #ifdef HAS_QUUX
472 foo = quux();
473 #endif
474
475 But in any case, try to keep the features and operating systems
476 separate.
477
478 Problematic System Interfaces
479 · malloc(0), realloc(0), calloc(0, 0) are non-portable. To be
480 portable allocate at least one byte. (In general you should rarely
481 need to work at this low level, but instead use the various malloc
482 wrappers.)
483
484 · snprintf() - the return type is unportable. Use my_snprintf()
485 instead.
486
487 Security problems
488 Last but not least, here are various tips for safer coding.
489
490 · Do not use gets()
491
492 Or we will publicly ridicule you. Seriously.
493
494 · Do not use strcpy() or strcat() or strncpy() or strncat()
495
496 Use my_strlcpy() and my_strlcat() instead: they either use the
497 native implementation, or Perl's own implementation (borrowed from
498 the public domain implementation of INN).
499
500 · Do not use sprintf() or vsprintf()
501
502 If you really want just plain byte strings, use my_snprintf() and
503 my_vsnprintf() instead, which will try to use snprintf() and
504 vsnprintf() if those safer APIs are available. If you want
505 something fancier than a plain byte string, use SVs and
506 Perl_sv_catpvf().
507
509 You can compile a special debugging version of Perl, which allows you
510 to use the "-D" option of Perl to tell more about what Perl is doing.
511 But sometimes there is no alternative than to dive in with a debugger,
512 either to see the stack trace of a core dump (very useful in a bug
513 report), or trying to figure out what went wrong before the core dump
514 happened, or how did we end up having wrong or unexpected results.
515
516 Poking at Perl
517 To really poke around with Perl, you'll probably want to build Perl for
518 debugging, like this:
519
520 ./Configure -d -D optimize=-g
521 make
522
523 "-g" is a flag to the C compiler to have it produce debugging
524 information which will allow us to step through a running program, and
525 to see in which C function we are at (without the debugging information
526 we might see only the numerical addresses of the functions, which is
527 not very helpful).
528
529 Configure will also turn on the "DEBUGGING" compilation symbol which
530 enables all the internal debugging code in Perl. There are a whole
531 bunch of things you can debug with this: perlrun lists them all, and
532 the best way to find out about them is to play about with them. The
533 most useful options are probably
534
535 l Context (loop) stack processing
536 t Trace execution
537 o Method and overloading resolution
538 c String/numeric conversions
539
540 Some of the functionality of the debugging code can be achieved using
541 XS modules.
542
543 -Dr => use re 'debug'
544 -Dx => use O 'Debug'
545
546 Using a source-level debugger
547 If the debugging output of "-D" doesn't help you, it's time to step
548 through perl's execution with a source-level debugger.
549
550 · We'll use "gdb" for our examples here; the principles will apply to
551 any debugger (many vendors call their debugger "dbx"), but check the
552 manual of the one you're using.
553
554 To fire up the debugger, type
555
556 gdb ./perl
557
558 Or if you have a core dump:
559
560 gdb ./perl core
561
562 You'll want to do that in your Perl source tree so the debugger can
563 read the source code. You should see the copyright message, followed by
564 the prompt.
565
566 (gdb)
567
568 "help" will get you into the documentation, but here are the most
569 useful commands:
570
571 · run [args]
572
573 Run the program with the given arguments.
574
575 · break function_name
576
577 · break source.c:xxx
578
579 Tells the debugger that we'll want to pause execution when we reach
580 either the named function (but see "Internal Functions" in
581 perlguts!) or the given line in the named source file.
582
583 · step
584
585 Steps through the program a line at a time.
586
587 · next
588
589 Steps through the program a line at a time, without descending into
590 functions.
591
592 · continue
593
594 Run until the next breakpoint.
595
596 · finish
597
598 Run until the end of the current function, then stop again.
599
600 · 'enter'
601
602 Just pressing Enter will do the most recent operation again - it's a
603 blessing when stepping through miles of source code.
604
605 · print
606
607 Execute the given C code and print its results. WARNING: Perl makes
608 heavy use of macros, and gdb does not necessarily support macros
609 (see later "gdb macro support"). You'll have to substitute them
610 yourself, or to invoke cpp on the source code files (see "The .i
611 Targets") So, for instance, you can't say
612
613 print SvPV_nolen(sv)
614
615 but you have to say
616
617 print Perl_sv_2pv_nolen(sv)
618
619 You may find it helpful to have a "macro dictionary", which you can
620 produce by saying "cpp -dM perl.c | sort". Even then, cpp won't
621 recursively apply those macros for you.
622
623 gdb macro support
624 Recent versions of gdb have fairly good macro support, but in order to
625 use it you'll need to compile perl with macro definitions included in
626 the debugging information. Using gcc version 3.1, this means
627 configuring with "-Doptimize=-g3". Other compilers might use a
628 different switch (if they support debugging macros at all).
629
630 Dumping Perl Data Structures
631 One way to get around this macro hell is to use the dumping functions
632 in dump.c; these work a little like an internal Devel::Peek, but they
633 also cover OPs and other structures that you can't get at from Perl.
634 Let's take an example. We'll use the "$a = $b + $c" we used before,
635 but give it a bit of context: "$b = "6XXXX"; $c = 2.3;". Where's a good
636 place to stop and poke around?
637
638 What about "pp_add", the function we examined earlier to implement the
639 "+" operator:
640
641 (gdb) break Perl_pp_add
642 Breakpoint 1 at 0x46249f: file pp_hot.c, line 309.
643
644 Notice we use "Perl_pp_add" and not "pp_add" - see "Internal Functions"
645 in perlguts. With the breakpoint in place, we can run our program:
646
647 (gdb) run -e '$b = "6XXXX"; $c = 2.3; $a = $b + $c'
648
649 Lots of junk will go past as gdb reads in the relevant source files and
650 libraries, and then:
651
652 Breakpoint 1, Perl_pp_add () at pp_hot.c:309
653 309 dSP; dATARGET; tryAMAGICbin(add,opASSIGN);
654 (gdb) step
655 311 dPOPTOPnnrl_ul;
656 (gdb)
657
658 We looked at this bit of code before, and we said that "dPOPTOPnnrl_ul"
659 arranges for two "NV"s to be placed into "left" and "right" - let's
660 slightly expand it:
661
662 #define dPOPTOPnnrl_ul NV right = POPn; \
663 SV *leftsv = TOPs; \
664 NV left = USE_LEFT(leftsv) ? SvNV(leftsv) : 0.0
665
666 "POPn" takes the SV from the top of the stack and obtains its NV either
667 directly (if "SvNOK" is set) or by calling the "sv_2nv" function.
668 "TOPs" takes the next SV from the top of the stack - yes, "POPn" uses
669 "TOPs" - but doesn't remove it. We then use "SvNV" to get the NV from
670 "leftsv" in the same way as before - yes, "POPn" uses "SvNV".
671
672 Since we don't have an NV for $b, we'll have to use "sv_2nv" to convert
673 it. If we step again, we'll find ourselves there:
674
675 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1669
676 1669 if (!sv)
677 (gdb)
678
679 We can now use "Perl_sv_dump" to investigate the SV:
680
681 SV = PV(0xa057cc0) at 0xa0675d0
682 REFCNT = 1
683 FLAGS = (POK,pPOK)
684 PV = 0xa06a510 "6XXXX"\0
685 CUR = 5
686 LEN = 6
687 $1 = void
688
689 We know we're going to get 6 from this, so let's finish the subroutine:
690
691 (gdb) finish
692 Run till exit from #0 Perl_sv_2nv (sv=0xa0675d0) at sv.c:1671
693 0x462669 in Perl_pp_add () at pp_hot.c:311
694 311 dPOPTOPnnrl_ul;
695
696 We can also dump out this op: the current op is always stored in
697 "PL_op", and we can dump it with "Perl_op_dump". This'll give us
698 similar output to B::Debug.
699
700 {
701 13 TYPE = add ===> 14
702 TARG = 1
703 FLAGS = (SCALAR,KIDS)
704 {
705 TYPE = null ===> (12)
706 (was rv2sv)
707 FLAGS = (SCALAR,KIDS)
708 {
709 11 TYPE = gvsv ===> 12
710 FLAGS = (SCALAR)
711 GV = main::b
712 }
713 }
714
715 # finish this later #
716
718 Various tools exist for analysing C source code statically, as opposed
719 to dynamically, that is, without executing the code. It is possible to
720 detect resource leaks, undefined behaviour, type mismatches,
721 portability problems, code paths that would cause illegal memory
722 accesses, and other similar problems by just parsing the C code and
723 looking at the resulting graph, what does it tell about the execution
724 and data flows. As a matter of fact, this is exactly how C compilers
725 know to give warnings about dubious code.
726
727 lint, splint
728 The good old C code quality inspector, "lint", is available in several
729 platforms, but please be aware that there are several different
730 implementations of it by different vendors, which means that the flags
731 are not identical across different platforms.
732
733 There is a lint variant called "splint" (Secure Programming Lint)
734 available from http://www.splint.org/ that should compile on any Unix-
735 like platform.
736
737 There are "lint" and <splint> targets in Makefile, but you may have to
738 diddle with the flags (see above).
739
740 Coverity
741 Coverity (http://www.coverity.com/) is a product similar to lint and as
742 a testbed for their product they periodically check several open source
743 projects, and they give out accounts to open source developers to the
744 defect databases.
745
746 cpd (cut-and-paste detector)
747 The cpd tool detects cut-and-paste coding. If one instance of the cut-
748 and-pasted code changes, all the other spots should probably be
749 changed, too. Therefore such code should probably be turned into a
750 subroutine or a macro.
751
752 cpd (http://pmd.sourceforge.net/cpd.html) is part of the pmd project
753 (http://pmd.sourceforge.net/). pmd was originally written for static
754 analysis of Java code, but later the cpd part of it was extended to
755 parse also C and C++.
756
757 Download the pmd-bin-X.Y.zip () from the SourceForge site, extract the
758 pmd-X.Y.jar from it, and then run that on source code thusly:
759
760 java -cp pmd-X.Y.jar net.sourceforge.pmd.cpd.CPD --minimum-tokens 100 --files /some/where/src --language c > cpd.txt
761
762 You may run into memory limits, in which case you should use the -Xmx
763 option:
764
765 java -Xmx512M ...
766
767 gcc warnings
768 Though much can be written about the inconsistency and coverage
769 problems of gcc warnings (like "-Wall" not meaning "all the warnings",
770 or some common portability problems not being covered by "-Wall", or
771 "-ansi" and "-pedantic" both being a poorly defined collection of
772 warnings, and so forth), gcc is still a useful tool in keeping our
773 coding nose clean.
774
775 The "-Wall" is by default on.
776
777 The "-ansi" (and its sidekick, "-pedantic") would be nice to be on
778 always, but unfortunately they are not safe on all platforms, they can
779 for example cause fatal conflicts with the system headers (Solaris
780 being a prime example). If Configure "-Dgccansipedantic" is used, the
781 "cflags" frontend selects "-ansi -pedantic" for the platforms where
782 they are known to be safe.
783
784 Starting from Perl 5.9.4 the following extra flags are added:
785
786 · "-Wendif-labels"
787
788 · "-Wextra"
789
790 · "-Wdeclaration-after-statement"
791
792 The following flags would be nice to have but they would first need
793 their own Augean stablemaster:
794
795 · "-Wpointer-arith"
796
797 · "-Wshadow"
798
799 · "-Wstrict-prototypes"
800
801 The "-Wtraditional" is another example of the annoying tendency of gcc
802 to bundle a lot of warnings under one switch (it would be impossible to
803 deploy in practice because it would complain a lot) but it does contain
804 some warnings that would be beneficial to have available on their own,
805 such as the warning about string constants inside macros containing the
806 macro arguments: this behaved differently pre-ANSI than it does in
807 ANSI, and some C compilers are still in transition, AIX being an
808 example.
809
810 Warnings of other C compilers
811 Other C compilers (yes, there are other C compilers than gcc) often
812 have their "strict ANSI" or "strict ANSI with some portability
813 extensions" modes on, like for example the Sun Workshop has its "-Xa"
814 mode on (though implicitly), or the DEC (these days, HP...) has its
815 "-std1" mode on.
816
818 NOTE 1: Running under memory debuggers such as Purify, valgrind, or
819 Third Degree greatly slows down the execution: seconds become minutes,
820 minutes become hours. For example as of Perl 5.8.1, the
821 ext/Encode/t/Unicode.t takes extraordinarily long to complete under
822 e.g. Purify, Third Degree, and valgrind. Under valgrind it takes more
823 than six hours, even on a snappy computer. The said test must be doing
824 something that is quite unfriendly for memory debuggers. If you don't
825 feel like waiting, that you can simply kill away the perl process.
826
827 NOTE 2: To minimize the number of memory leak false alarms (see
828 "PERL_DESTRUCT_LEVEL" for more information), you have to set the
829 environment variable PERL_DESTRUCT_LEVEL to 2.
830
831 For csh-like shells:
832
833 setenv PERL_DESTRUCT_LEVEL 2
834
835 For Bourne-type shells:
836
837 PERL_DESTRUCT_LEVEL=2
838 export PERL_DESTRUCT_LEVEL
839
840 In Unixy environments you can also use the "env" command:
841
842 env PERL_DESTRUCT_LEVEL=2 valgrind ./perl -Ilib ...
843
844 NOTE 3: There are known memory leaks when there are compile-time errors
845 within eval or require, seeing "S_doeval" in the call stack is a good
846 sign of these. Fixing these leaks is non-trivial, unfortunately, but
847 they must be fixed eventually.
848
849 NOTE 4: DynaLoader will not clean up after itself completely unless
850 Perl is built with the Configure option
851 "-Accflags=-DDL_UNLOAD_ALL_AT_EXIT".
852
853 Rational Software's Purify
854 Purify is a commercial tool that is helpful in identifying memory
855 overruns, wild pointers, memory leaks and other such badness. Perl must
856 be compiled in a specific way for optimal testing with Purify. Purify
857 is available under Windows NT, Solaris, HP-UX, SGI, and Siemens Unix.
858
859 Purify on Unix
860
861 On Unix, Purify creates a new Perl binary. To get the most benefit out
862 of Purify, you should create the perl to Purify using:
863
864 sh Configure -Accflags=-DPURIFY -Doptimize='-g' \
865 -Uusemymalloc -Dusemultiplicity
866
867 where these arguments mean:
868
869 · -Accflags=-DPURIFY
870
871 Disables Perl's arena memory allocation functions, as well as
872 forcing use of memory allocation functions derived from the system
873 malloc.
874
875 · -Doptimize='-g'
876
877 Adds debugging information so that you see the exact source
878 statements where the problem occurs. Without this flag, all you
879 will see is the source filename of where the error occurred.
880
881 · -Uusemymalloc
882
883 Disable Perl's malloc so that Purify can more closely monitor
884 allocations and leaks. Using Perl's malloc will make Purify report
885 most leaks in the "potential" leaks category.
886
887 · -Dusemultiplicity
888
889 Enabling the multiplicity option allows perl to clean up thoroughly
890 when the interpreter shuts down, which reduces the number of bogus
891 leak reports from Purify.
892
893 Once you've compiled a perl suitable for Purify'ing, then you can just:
894
895 make pureperl
896
897 which creates a binary named 'pureperl' that has been Purify'ed. This
898 binary is used in place of the standard 'perl' binary when you want to
899 debug Perl memory problems.
900
901 As an example, to show any memory leaks produced during the standard
902 Perl testset you would create and run the Purify'ed perl as:
903
904 make pureperl
905 cd t
906 ../pureperl -I../lib harness
907
908 which would run Perl on test.pl and report any memory problems.
909
910 Purify outputs messages in "Viewer" windows by default. If you don't
911 have a windowing environment or if you simply want the Purify output to
912 unobtrusively go to a log file instead of to the interactive window,
913 use these following options to output to the log file "perl.log":
914
915 setenv PURIFYOPTIONS "-chain-length=25 -windows=no \
916 -log-file=perl.log -append-logfile=yes"
917
918 If you plan to use the "Viewer" windows, then you only need this
919 option:
920
921 setenv PURIFYOPTIONS "-chain-length=25"
922
923 In Bourne-type shells:
924
925 PURIFYOPTIONS="..."
926 export PURIFYOPTIONS
927
928 or if you have the "env" utility:
929
930 env PURIFYOPTIONS="..." ../pureperl ...
931
932 Purify on NT
933
934 Purify on Windows NT instruments the Perl binary 'perl.exe' on the fly.
935 There are several options in the makefile you should change to get the
936 most use out of Purify:
937
938 · DEFINES
939
940 You should add -DPURIFY to the DEFINES line so the DEFINES line
941 looks something like:
942
943 DEFINES = -DWIN32 -D_CONSOLE -DNO_STRICT $(CRYPT_FLAG) -DPURIFY=1
944
945 to disable Perl's arena memory allocation functions, as well as to
946 force use of memory allocation functions derived from the system
947 malloc.
948
949 · USE_MULTI = define
950
951 Enabling the multiplicity option allows perl to clean up thoroughly
952 when the interpreter shuts down, which reduces the number of bogus
953 leak reports from Purify.
954
955 · #PERL_MALLOC = define
956
957 Disable Perl's malloc so that Purify can more closely monitor
958 allocations and leaks. Using Perl's malloc will make Purify report
959 most leaks in the "potential" leaks category.
960
961 · CFG = Debug
962
963 Adds debugging information so that you see the exact source
964 statements where the problem occurs. Without this flag, all you
965 will see is the source filename of where the error occurred.
966
967 As an example, to show any memory leaks produced during the standard
968 Perl testset you would create and run Purify as:
969
970 cd win32
971 make
972 cd ../t
973 purify ../perl -I../lib harness
974
975 which would instrument Perl in memory, run Perl on test.pl, then
976 finally report any memory problems.
977
978 valgrind
979 The excellent valgrind tool can be used to find out both memory leaks
980 and illegal memory accesses. As of version 3.3.0, Valgrind only
981 supports Linux on x86, x86-64 and PowerPC and Darwin (OS X) on x86 and
982 x86-64). The special "test.valgrind" target can be used to run the
983 tests under valgrind. Found errors and memory leaks are logged in files
984 named testfile.valgrind.
985
986 Valgrind also provides a cachegrind tool, invoked on perl as:
987
988 VG_OPTS=--tool=cachegrind make test.valgrind
989
990 As system libraries (most notably glibc) are also triggering errors,
991 valgrind allows to suppress such errors using suppression files. The
992 default suppression file that comes with valgrind already catches a lot
993 of them. Some additional suppressions are defined in t/perl.supp.
994
995 To get valgrind and for more information see
996
997 http://valgrind.org/
998
1000 Depending on your platform there are various ways of profiling Perl.
1001
1002 There are two commonly used techniques of profiling executables:
1003 statistical time-sampling and basic-block counting.
1004
1005 The first method takes periodically samples of the CPU program counter,
1006 and since the program counter can be correlated with the code generated
1007 for functions, we get a statistical view of in which functions the
1008 program is spending its time. The caveats are that very small/fast
1009 functions have lower probability of showing up in the profile, and that
1010 periodically interrupting the program (this is usually done rather
1011 frequently, in the scale of milliseconds) imposes an additional
1012 overhead that may skew the results. The first problem can be alleviated
1013 by running the code for longer (in general this is a good idea for
1014 profiling), the second problem is usually kept in guard by the
1015 profiling tools themselves.
1016
1017 The second method divides up the generated code into basic blocks.
1018 Basic blocks are sections of code that are entered only in the
1019 beginning and exited only at the end. For example, a conditional jump
1020 starts a basic block. Basic block profiling usually works by
1021 instrumenting the code by adding enter basic block #nnnn book-keeping
1022 code to the generated code. During the execution of the code the basic
1023 block counters are then updated appropriately. The caveat is that the
1024 added extra code can skew the results: again, the profiling tools
1025 usually try to factor their own effects out of the results.
1026
1027 Gprof Profiling
1028 gprof is a profiling tool available in many Unix platforms, it uses
1029 statistical time-sampling.
1030
1031 You can build a profiled version of perl called "perl.gprof" by
1032 invoking the make target "perl.gprof" (What is required is that Perl
1033 must be compiled using the "-pg" flag, you may need to re-Configure).
1034 Running the profiled version of Perl will create an output file called
1035 gmon.out is created which contains the profiling data collected during
1036 the execution.
1037
1038 The gprof tool can then display the collected data in various ways.
1039 Usually gprof understands the following options:
1040
1041 · -a
1042
1043 Suppress statically defined functions from the profile.
1044
1045 · -b
1046
1047 Suppress the verbose descriptions in the profile.
1048
1049 · -e routine
1050
1051 Exclude the given routine and its descendants from the profile.
1052
1053 · -f routine
1054
1055 Display only the given routine and its descendants in the profile.
1056
1057 · -s
1058
1059 Generate a summary file called gmon.sum which then may be given to
1060 subsequent gprof runs to accumulate data over several runs.
1061
1062 · -z
1063
1064 Display routines that have zero usage.
1065
1066 For more detailed explanation of the available commands and output
1067 formats, see your own local documentation of gprof.
1068
1069 quick hint:
1070
1071 $ sh Configure -des -Dusedevel -Doptimize='-pg' && make perl.gprof
1072 $ ./perl.gprof someprog # creates gmon.out in current directory
1073 $ gprof ./perl.gprof > out
1074 $ view out
1075
1076 GCC gcov Profiling
1077 Starting from GCC 3.0 basic block profiling is officially available for
1078 the GNU CC.
1079
1080 You can build a profiled version of perl called perl.gcov by invoking
1081 the make target "perl.gcov" (what is required that Perl must be
1082 compiled using gcc with the flags "-fprofile-arcs -ftest-coverage", you
1083 may need to re-Configure).
1084
1085 Running the profiled version of Perl will cause profile output to be
1086 generated. For each source file an accompanying ".da" file will be
1087 created.
1088
1089 To display the results you use the "gcov" utility (which should be
1090 installed if you have gcc 3.0 or newer installed). gcov is run on
1091 source code files, like this
1092
1093 gcov sv.c
1094
1095 which will cause sv.c.gcov to be created. The .gcov files contain the
1096 source code annotated with relative frequencies of execution indicated
1097 by "#" markers.
1098
1099 Useful options of gcov include "-b" which will summarise the basic
1100 block, branch, and function call coverage, and "-c" which instead of
1101 relative frequencies will use the actual counts. For more information
1102 on the use of gcov and basic block profiling with gcc, see the latest
1103 GNU CC manual, as of GCC 3.0 see
1104
1105 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc.html
1106
1107 and its section titled "8. gcov: a Test Coverage Program"
1108
1109 http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc_8.html#SEC132
1110
1111 quick hint:
1112
1113 $ sh Configure -des -Dusedevel -Doptimize='-g' \
1114 -Accflags='-fprofile-arcs -ftest-coverage' \
1115 -Aldflags='-fprofile-arcs -ftest-coverage' && make perl.gcov
1116 $ rm -f regexec.c.gcov regexec.gcda
1117 $ ./perl.gcov
1118 $ gcov regexec.c
1119 $ view regexec.c.gcov
1120
1122 PERL_DESTRUCT_LEVEL
1123 If you want to run any of the tests yourself manually using e.g.
1124 valgrind, or the pureperl or perl.third executables, please note that
1125 by default perl does not explicitly cleanup all the memory it has
1126 allocated (such as global memory arenas) but instead lets the exit() of
1127 the whole program "take care" of such allocations, also known as
1128 "global destruction of objects".
1129
1130 There is a way to tell perl to do complete cleanup: set the environment
1131 variable PERL_DESTRUCT_LEVEL to a non-zero value. The t/TEST wrapper
1132 does set this to 2, and this is what you need to do too, if you don't
1133 want to see the "global leaks": For example, for "third-degreed" Perl:
1134
1135 env PERL_DESTRUCT_LEVEL=2 ./perl.third -Ilib t/foo/bar.t
1136
1137 (Note: the mod_perl apache module uses also this environment variable
1138 for its own purposes and extended its semantics. Refer to the mod_perl
1139 documentation for more information. Also, spawned threads do the
1140 equivalent of setting this variable to the value 1.)
1141
1142 If, at the end of a run you get the message N scalars leaked, you can
1143 recompile with "-DDEBUG_LEAKING_SCALARS", which will cause the
1144 addresses of all those leaked SVs to be dumped along with details as to
1145 where each SV was originally allocated. This information is also
1146 displayed by Devel::Peek. Note that the extra details recorded with
1147 each SV increases memory usage, so it shouldn't be used in production
1148 environments. It also converts "new_SV()" from a macro into a real
1149 function, so you can use your favourite debugger to discover where
1150 those pesky SVs were allocated.
1151
1152 If you see that you're leaking memory at runtime, but neither valgrind
1153 nor "-DDEBUG_LEAKING_SCALARS" will find anything, you're probably
1154 leaking SVs that are still reachable and will be properly cleaned up
1155 during destruction of the interpreter. In such cases, using the "-Dm"
1156 switch can point you to the source of the leak. If the executable was
1157 built with "-DDEBUG_LEAKING_SCALARS", "-Dm" will output SV allocations
1158 in addition to memory allocations. Each SV allocation has a distinct
1159 serial number that will be written on creation and destruction of the
1160 SV. So if you're executing the leaking code in a loop, you need to look
1161 for SVs that are created, but never destroyed between each cycle. If
1162 such an SV is found, set a conditional breakpoint within "new_SV()" and
1163 make it break only when "PL_sv_serial" is equal to the serial number of
1164 the leaking SV. Then you will catch the interpreter in exactly the
1165 state where the leaking SV is allocated, which is sufficient in many
1166 cases to find the source of the leak.
1167
1168 As "-Dm" is using the PerlIO layer for output, it will by itself
1169 allocate quite a bunch of SVs, which are hidden to avoid recursion. You
1170 can bypass the PerlIO layer if you use the SV logging provided by
1171 "-DPERL_MEM_LOG" instead.
1172
1173 PERL_MEM_LOG
1174 If compiled with "-DPERL_MEM_LOG", both memory and SV allocations go
1175 through logging functions, which is handy for breakpoint setting.
1176
1177 Unless "-DPERL_MEM_LOG_NOIMPL" is also compiled, the logging functions
1178 read $ENV{PERL_MEM_LOG} to determine whether to log the event, and if
1179 so how:
1180
1181 $ENV{PERL_MEM_LOG} =~ /m/ Log all memory ops
1182 $ENV{PERL_MEM_LOG} =~ /s/ Log all SV ops
1183 $ENV{PERL_MEM_LOG} =~ /t/ include timestamp in Log
1184 $ENV{PERL_MEM_LOG} =~ /^(\d+)/ write to FD given (default is 2)
1185
1186 Memory logging is somewhat similar to "-Dm" but is independent of
1187 "-DDEBUGGING", and at a higher level; all uses of Newx(), Renew(), and
1188 Safefree() are logged with the caller's source code file and line
1189 number (and C function name, if supported by the C compiler). In
1190 contrast, "-Dm" is directly at the point of "malloc()". SV logging is
1191 similar.
1192
1193 Since the logging doesn't use PerlIO, all SV allocations are logged and
1194 no extra SV allocations are introduced by enabling the logging. If
1195 compiled with "-DDEBUG_LEAKING_SCALARS", the serial number for each SV
1196 allocation is also logged.
1197
1198 DDD over gdb
1199 Those debugging perl with the DDD frontend over gdb may find the
1200 following useful:
1201
1202 You can extend the data conversion shortcuts menu, so for example you
1203 can display an SV's IV value with one click, without doing any typing.
1204 To do that simply edit ~/.ddd/init file and add after:
1205
1206 ! Display shortcuts.
1207 Ddd*gdbDisplayShortcuts: \
1208 /t () // Convert to Bin\n\
1209 /d () // Convert to Dec\n\
1210 /x () // Convert to Hex\n\
1211 /o () // Convert to Oct(\n\
1212
1213 the following two lines:
1214
1215 ((XPV*) (())->sv_any )->xpv_pv // 2pvx\n\
1216 ((XPVIV*) (())->sv_any )->xiv_iv // 2ivx
1217
1218 so now you can do ivx and pvx lookups or you can plug there the sv_peek
1219 "conversion":
1220
1221 Perl_sv_peek(my_perl, (SV*)()) // sv_peek
1222
1223 (The my_perl is for threaded builds.) Just remember that every line,
1224 but the last one, should end with \n\
1225
1226 Alternatively edit the init file interactively via: 3rd mouse button ->
1227 New Display -> Edit Menu
1228
1229 Note: you can define up to 20 conversion shortcuts in the gdb section.
1230
1231 Poison
1232 If you see in a debugger a memory area mysteriously full of 0xABABABAB
1233 or 0xEFEFEFEF, you may be seeing the effect of the Poison() macros, see
1234 perlclib.
1235
1236 Read-only optrees
1237 Under ithreads the optree is read only. If you want to enforce this, to
1238 check for write accesses from buggy code, compile with
1239 "-DPL_OP_SLAB_ALLOC" to enable the OP slab allocator and
1240 "-DPERL_DEBUG_READONLY_OPS" to enable code that allocates op memory via
1241 "mmap", and sets it read-only at run time. Any write access to an op
1242 results in a "SIGBUS" and abort.
1243
1244 This code is intended for development only, and may not be portable
1245 even to all Unix variants. Also, it is an 80% solution, in that it
1246 isn't able to make all ops read only. Specifically it
1247
1248 · 1
1249
1250 Only sets read-only on all slabs of ops at "CHECK" time, hence ops
1251 allocated later via "require" or "eval" will be re-write
1252
1253 · 2
1254
1255 Turns an entire slab of ops read-write if the refcount of any op in
1256 the slab needs to be decreased.
1257
1258 · 3
1259
1260 Turns an entire slab of ops read-write if any op from the slab is
1261 freed.
1262
1263 It's not possible to turn the slabs to read-only after an action
1264 requiring read-write access, as either can happen during op tree
1265 building time, so there may still be legitimate write access.
1266
1267 However, as an 80% solution it is still effective, as currently it
1268 catches a write access during the generation of Config.pm, which means
1269 that we can't yet build perl with this enabled.
1270
1271 The .i Targets
1272 You can expand the macros in a foo.c file by saying
1273
1274 make foo.i
1275
1276 which will expand the macros using cpp. Don't be scared by the
1277 results.
1278
1280 This document was originally written by Nathan Torkington, and is
1281 maintained by the perl5-porters mailing list.
1282
1283
1284
1285perl v5.16.3 2013-03-04 PERLHACKTIPS(1)