3 # Try opening libperl.a with nm, and verifying it has the kind of
4 # symbols we expect, and no symbols we should avoid.
6 # Fail softly, expect things only on known platforms:
8 # - darwin (OS X), both x86 and ppc
10 # and on other platforms, and if things seem odd, just give up (skip_all).
12 # Also, if the rarely-used builds options -DPERL_GLOBAL_STRUCT or
13 # -DPERL_GLOBAL_STRUCT_PRIVATE are used, verify that they did what
14 # they were meant to do, hide the global variables (see perlguts for
17 # Debugging tip: nm output (this script's input) can be faked by
18 # giving one command line argument for this script: it should be
19 # either the filename to read, or "-" for STDIN. You can also append
20 # "@style" (where style is a supported nm style, like "gnu" or "darwin")
21 # to this filename for "cross-parsing".
24 # - "text" symbols are code
25 # - "data" symbols are data (duh), with subdivisions:
26 # - "bss": (Block-Started-by-Symbol: originally from IBM assembler...),
27 # uninitialized data, which often even doesn't exist in the object
28 # file as such, only its size does, which is then created on demand
30 # - "const": initialized read-only data, like string literals
31 # - "common": uninitialized data unless initialized...
32 # (the full story is too long for here, see "man nm")
33 # - "data": initialized read-write data
34 # (somewhat confusingly below: "data data", but it makes code simpler)
35 # - "undefined": external symbol referred to by an object,
36 # most likely a text symbol. Can be either a symbol defined by
37 # a Perl object file but referred to by other Perl object files,
38 # or a completely external symbol from libc, or other system libraries.
50 if ($Config{cc} =~ /g\+\+/) {
51 # XXX Could use c++filt, maybe.
57 for my $f (qw(../libperl.a libperl.a)) {
64 unless (defined $libperl_a) {
65 skip_all "no libperl.a";
68 print "# \$^O = $^O\n";
69 print "# \$Config{cc} = $Config{cc}\n";
70 print "# libperl = $libperl_a\n";
76 my $nm_err_tmp = "libperl$$";
79 # this is still executed when we skip_all above, avoid a warning
80 unlink $nm_err_tmp if $nm_err_tmp;
87 $fake_input = shift @ARGV;
88 print "# Faking nm output from $fake_input\n";
89 if ($fake_input =~ s/\@(.+)$//) {
91 print "# Faking nm style from $fake_style\n";
92 if ($fake_style eq 'gnu' ||
93 $fake_style eq 'linux' ||
94 $fake_style eq 'freebsd') {
96 } elsif ($fake_style eq 'darwin' || $fake_style eq 'osx') {
99 die "$0: Unknown explicit nm style '$fake_style'\n";
104 unless (defined $nm_style) {
105 if ($^O eq 'linux') {
106 # The 'gnu' style could be equally well be called 'bsd' style,
107 # since the output format of the GNU binutils nm is really BSD.
109 } elsif ($^O eq 'freebsd') {
111 } elsif ($^O eq 'darwin') {
112 $nm_style = 'darwin';
116 if (defined $nm_style) {
117 if ($nm_style eq 'gnu') {
119 } elsif ($nm_style eq 'darwin') {
121 # With the -m option we get better information than the BSD-like
122 # default: with the default, a lot of symbols get dumped into 'S'
123 # or 's', for example one cannot tell the difference between const
124 # and non-const data symbols.
127 die "$0: Unexpected nm style '$nm_style'\n";
131 unless (defined $nm) {
135 unless (defined $nm_style) {
136 skip_all "no nm style";
139 print "# nm = $nm\n";
140 print "# nm_style = $nm_style\n";
141 print "# nm_opt = $nm_opt\n";
144 skip_all "no executable nm $nm";
147 if ($nm_style eq 'gnu' && !defined $fake_style) {
148 open(my $gnu_verify, "$nm --version|") or
149 skip_all "nm failed: $!";
151 while (<$gnu_verify>) {
157 unless ($gnu_verified) {
158 skip_all "no GNU nm";
162 if (defined $fake_input) {
163 if ($fake_input eq '-') {
164 open($nm_fh, "<&STDIN") or
165 skip_all "Duping STDIN failed: $!";
167 open($nm_fh, "<", $fake_input) or
168 skip_all "Opening '$fake_input' failed: $!";
170 undef $nm_err_tmp; # In this case there will be no nm errors.
172 open($nm_fh, "$nm $nm_opt $libperl_a 2>$nm_err_tmp |") or
173 skip_all "$nm $nm_opt $libperl_a failed: $!";
176 sub is_perlish_symbol {
177 $_[0] =~ /^(?:PL_|Perl|PerlIO)/;
180 # XXX Implement "internal test" for this script (option -t?)
181 # to verify that the parsing does what it's intended to.
186 if (m{^(\w+\.o):$}) {
188 $symbols->{obj}{$1}++;
192 die "$0: undefined current object: $line"
193 unless defined $symbols->{o};
194 # 64-bit systems have 16 hexdigits, 32-bit systems have 8.
195 if (s/^[0-9a-f]{8}(?:[0-9a-f]{8})? //) {
196 if (/^[Rr] (\w+)$/) {
197 # R: read only (const)
198 $symbols->{data}{const}{$1}{$symbols->{o}}++;
200 # Skip local const (read only).
201 } elsif (/^[Tti] (\w+)(\..+)?$/) {
202 $symbols->{text}{$1}{$symbols->{o}}++;
203 } elsif (/^C (\w+)$/) {
204 $symbols->{data}{common}{$1}{$symbols->{o}}++;
205 } elsif (/^[BbSs] (\w+)(\.\d+)?$/) {
206 # Bb: uninitialized data (bss)
207 # Ss: uninitialized data "for small objects"
208 $symbols->{data}{bss}{$1}{$symbols->{o}}++;
209 } elsif (/^D _LIB_VERSION$/) {
210 # Skip the _LIB_VERSION (not ours, probably libm)
211 } elsif (/^[DdGg] (\w+)$/) {
212 # Dd: initialized data
213 # Gg: initialized "for small objects"
214 $symbols->{data}{data}{$1}{$symbols->{o}}++;
215 } elsif (/^. \.?(\w+)$/) {
216 # Skip the unknown types.
217 print "# Unknown type: $line ($symbols->{o})\n";
220 } elsif (/^ {8}(?: {8})? U _?(\w+)$/) {
222 return if is_perlish_symbol($symbol);
223 $symbols->{undef}{$symbol}{$symbols->{o}}++;
227 print "# Unexpected nm output '$line' ($symbols->{o})\n";
230 sub nm_parse_darwin {
233 if (m{^(?:.+)?libperl\.a\((\w+\.o)\):$}) {
235 $symbols->{obj}{$1}++;
239 die "$0: undefined current object: $line" unless defined $symbols->{o};
240 # 64-bit systems have 16 hexdigits, 32-bit systems have 8.
241 if (s/^[0-9a-f]{8}(?:[0-9a-f]{8})? //) {
242 # String literals can live in different sections
243 # depending on the compiler and os release, assumedly
245 if (/^\(__TEXT,__(?:const|cstring|literal\d+)\) (?:non-)?external _?(\w+)(\.\w+)?$/) {
246 my ($symbol, $suffix) = ($1, $2);
247 # Ignore function-local constants like
248 # _Perl_av_extend_guts.oom_array_extend
249 return if defined $suffix && /__TEXT,__const/;
250 # Ignore the cstring unnamed strings.
251 return if $symbol =~ /^L\.str\d+$/;
252 $symbols->{data}{const}{$symbol}{$symbols->{o}}++;
253 } elsif (/^\(__TEXT,__text\) (?:non-)?external _(\w+)$/) {
254 $symbols->{text}{$1}{$symbols->{o}}++;
255 } elsif (/^\(__DATA,__(const|data|bss|common)\) (?:non-)?external _(\w+)(\.\w+)?$/) {
256 my ($dtype, $symbol, $suffix) = ($1, $2, $3);
257 # Ignore function-local constants like
258 # _Perl_pp_gmtime.dayname
259 return if defined $suffix;
260 $symbols->{data}{$dtype}{$symbol}{$symbols->{o}}++;
261 } elsif (/^\(__DATA,__const\) non-external _\.memset_pattern\d*$/) {
262 # Skip this, whatever it is (some inlined leakage from
264 } elsif (/^\(__TEXT,__eh_frame/) {
265 # Skip the eh_frame (exception handling) symbols.
267 } elsif (/^\(__\w+,__\w+\) /) {
268 # Skip the unknown types.
269 print "# Unknown type: $line ($symbols->{o})\n";
272 } elsif (/^ {8}(?: {8})? \(undefined(?: \[lazy bound\])?\) external _?(.+)/) {
273 # darwin/ppc marks most undefined text symbols
276 return if is_perlish_symbol($symbol);
277 $symbols->{undef}{$symbol}{$symbols->{o}}++;
281 print "# Unexpected nm output '$line' ($symbols->{o})\n";
286 if ($nm_style eq 'gnu') {
287 $nm_parse = \&nm_parse_gnu;
288 } elsif ($nm_style eq 'darwin') {
289 $nm_parse = \&nm_parse_darwin;
292 unless (defined $nm_parse) {
293 skip_all "no nm parser ($nm_style $nm_style, \$^O $^O)";
301 $nm_parse->(\%symbols);
304 # use Data::Dumper; print Dumper(\%symbols);
306 if (keys %symbols == 0) {
307 skip_all "no symbols\n";
310 # These should always be true for everyone.
312 ok($symbols{obj}{'pp.o'}, "has object pp.o");
313 ok($symbols{text}{'Perl_peep'}, "has text Perl_peep");
314 ok($symbols{text}{'Perl_pp_uc'}{'pp.o'}, "has text Perl_pp_uc in pp.o");
315 ok(exists $symbols{data}{const}, "has data const symbols");
316 ok($symbols{data}{const}{PL_no_mem}{'globals.o'}, "has PL_no_mem");
318 my $DEBUGGING = $Config{ccflags} =~ /-DDEBUGGING/ ? 1 : 0;
320 my $GS = $Config{ccflags} =~ /-DPERL_GLOBAL_STRUCT\b/ ? 1 : 0;
321 my $GSP = $Config{ccflags} =~ /-DPERL_GLOBAL_STRUCT_PRIVATE/ ? 1 : 0;
323 print "# GS = $GS\n";
324 print "# GSP = $GSP\n";
328 for my $dtype (sort keys %{$symbols{data}}) {
329 for my $symbol (sort keys %{$symbols{data}{$dtype}}) {
330 $data_symbols{$symbol}++;
334 # The following tests differ between vanilla vs $GSP or $GS.
337 print "# -DPERL_GLOBAL_STRUCT_PRIVATE\n";
338 ok(!exists $data_symbols{PL_hash_seed}, "has no PL_hash_seed");
339 ok(!exists $data_symbols{PL_ppaddr}, "has no PL_ppaddr");
341 ok(! exists $symbols{data}{bss}, "has no data bss symbols");
342 ok(! exists $symbols{data}{data} ||
343 # clang with ASAN seems to add this symbol to every object file:
344 !grep($_ ne '__unnamed_1', keys %{$symbols{data}{data}}),
345 "has no data data symbols");
346 ok(! exists $symbols{data}{common}, "has no data common symbols");
348 # -DPERL_GLOBAL_STRUCT_PRIVATE should NOT have
349 # the extra text symbol for accessing the vars
350 # (as opposed to "just" -DPERL_GLOBAL_STRUCT)
351 ok(! exists $symbols{text}{Perl_GetVars}, "has no Perl_GetVars");
353 print "# -DPERL_GLOBAL_STRUCT\n";
354 ok(!exists $data_symbols{PL_hash_seed}, "has no PL_hash_seed");
355 ok(!exists $data_symbols{PL_ppaddr}, "has no PL_ppaddr");
357 ok(! exists $symbols{data}{bss}, "has no data bss symbols");
359 # These PerlIO data symbols are left visible with
360 # -DPERL_GLOBAL_STRUCT (as opposed to -DPERL_GLOBAL_STRUCT_PRIVATE)
374 # PL_magic_vtables is const with -DPERL_GLOBAL_STRUCT_PRIVATE but
375 # otherwise not const -- because of SWIG which wants to modify
376 # the table. Evil SWIG, eeevil.
378 # my_cxt_index is used with PERL_IMPLICIT_CONTEXT, which
379 # -DPERL_GLOBAL_STRUCT has turned on.
380 eq_array([sort keys %{$symbols{data}{data}}],
385 "data data symbols");
387 # Only one data common symbol, our "supervariable".
388 eq_array([sort keys %{$symbols{data}{common}}],
390 "data common symbols");
392 ok($symbols{data}{data}{PL_VarsPtr}{'globals.o'}, "has PL_VarsPtr");
393 ok($symbols{data}{common}{PL_Vars}{'globals.o'}, "has PL_Vars");
395 # -DPERL_GLOBAL_STRUCT has extra text symbol for accessing the vars.
396 ok($symbols{text}{Perl_GetVars}{'util.o'}, "has Perl_GetVars");
398 print "# neither -DPERL_GLOBAL_STRUCT nor -DPERL_GLOBAL_STRUCT_PRIVATE\n";
400 if ( !$symbols{data}{common} ) {
401 # This is likely because Perl was compiled with
402 # -Accflags="-fno-common"
403 $symbols{data}{common} = $symbols{data}{bss};
406 ok($symbols{data}{common}{PL_hash_seed}{'globals.o'}, "has PL_hash_seed");
407 ok($symbols{data}{data}{PL_ppaddr}{'globals.o'}, "has PL_ppaddr");
409 # None of the GLOBAL_STRUCT* business here.
410 ok(! exists $symbols{data}{data}{PL_VarsPtr}, "has no PL_VarsPtr");
411 ok(! exists $symbols{data}{common}{PL_Vars}, "has no PL_Vars");
412 ok(! exists $symbols{text}{Perl_GetVars}, "has no Perl_GetVars");
415 # See the comments in the beginning for what "undefined symbols"
416 # really means. We *should* have many of those, that is a good thing.
417 ok(keys %{$symbols{undef}}, "has undefined symbols");
419 # There are certain symbols we expect to see.
421 # chmod, socket, getenv, sigaction, sqrt, time are system/library
422 # calls that should each see at least one use. sqrt can be sqrtl
425 chmod => undef, # There is no Configure symbol for chmod.
426 socket => 'd_socket',
427 getenv => undef, # There is no Configure symbol for getenv,
428 sigaction => 'd_sigaction',
432 if ($Config{uselongdouble} && $Config{d_longdbl}) {
433 $expected{sqrtl} = 'd_sqrtl';
435 $expected{sqrt} = undef; # There is no Configure symbol for sqrt.
438 # DynaLoader will use dlopen, unless we are building static,
439 # and it is used in the platforms we are supporting in this test.
440 if ($Config{usedl} ) {
441 $expected{dlopen} = 'd_dlopen';
444 for my $symbol (sort keys %expected) {
445 if (defined $expected{$symbol} && !$Config{$expected{$symbol}}) {
451 my @o = exists $symbols{undef}{$symbol} ?
452 sort keys %{ $symbols{undef}{$symbol} } : ();
453 ok(@o, "uses $symbol (@o)");
456 # There are certain symbols we expect NOT to see.
458 # gets is horribly unsafe.
460 # fgets should not be used (Perl has its own API, sv_gets),
461 # even without perlio.
465 # strcat, strcpy, strncat, strncpy are unsafe.
467 # sprintf and vsprintf should not be used because
468 # Perl has its own safer and more portable implementations.
469 # (One exception: for certain floating point outputs
470 # the native sprintf is still used in some platforms, see below.)
472 # atoi has unsafe and undefined failure modes, and is affected by locale.
473 # Its cousins include atol and atoll.
475 # strtol and strtoul are affected by locale.
476 # Cousins include strtoq.
478 # system should not be used, use pp_system or my_popen.
483 for my $str (qw(system)) {
484 $unexpected{$str} = "d_$str";
487 for my $stdio (qw(gets fgets tmpfile sprintf vsprintf)) {
488 $unexpected{$stdio} = undef; # No Configure symbol for these.
490 for my $str (qw(strcat strcpy strncat strncpy)) {
491 $unexpected{$str} = undef; # No Configure symbol for these.
494 $unexpected{atoi} = undef; # No Configure symbol for atoi.
495 $unexpected{atol} = undef; # No Configure symbol for atol.
497 for my $str (qw(atoll strtol strtoul strtoq)) {
498 $unexpected{$str} = "d_$str";
501 for my $symbol (sort keys %unexpected) {
502 if (defined $unexpected{$symbol} && !$Config{$unexpected{$symbol}}) {
508 my @o = exists $symbols{undef}{$symbol} ?
509 sort keys %{ $symbols{undef}{$symbol} } : ();
510 # While sprintf() is bad in the general case,
511 # some platforms implement Gconvert via sprintf, in sv.o.
512 if ($symbol eq 'sprintf' &&
513 $Config{d_Gconvert} =~ /^sprintf/ &&
514 @o == 1 && $o[0] eq 'sv.o') {
516 skip("uses sprintf for Gconvert in sv.o");
519 is(@o, 0, "uses no $symbol (@o)");
523 if (defined $nm_err_tmp) {
524 if (open(my $nm_err_fh, $nm_err_tmp)) {
526 while (<$nm_err_fh>) {
527 # OS X has weird error where nm warns about
528 # "no name list" but then outputs fine.
529 if (/nm: no name list/ && $^O eq 'darwin') {
530 print "# $^O ignoring $nm output: $_";
533 warn "$0: Unexpected $nm error: $_";
536 die "$0: Unexpected $nm errors\n" if $error;
538 warn "Failed to open '$nm_err_tmp': $!\n";