在这里震耳欲聋的沉默之后,实际上我问的所有地方,我都转向了 valgrind-users@ML。这是线程:
http://sourceforge.net/mailarchive/forum.php?thread_name=e847e3a9-0d10-4c5e-929f-51258ecf9dfc%40iris&forum_name=valgrind-users
Josef 的回复非常有帮助,在#perl 的耐心帮助下,我编写了一个脚本,帮助我过滤掉调用图中不需要的库。
该脚本依赖于告诉 callgrind 更加冗长:
valgrind --tool=callgrind --dump-instr=yes --compress-pos=no \
--compress-strings=no /opt/ats-trunk/bin/traffic_server
这样,它将生成字符串而不是参考编号,从而更容易解析:
#!/usr/bin/perl
use Modern::Perl;
require File::Temp;
my $cob = qr{^cob=/(?:usr/)?lib};
my $ob = qr{^ob=/(?:usr/)?lib/};
my $calls = qr{^calls=};
open (my $fh, '<', $ARGV[0]) or die $!;
my $tmp = File::Temp->new(UNLINK => 1);
## Skip all external libraries, as defined by $ob
while (readline $fh) {
if (/$ob/ ) {
# skip the entire ob= section we don't need.
0 while defined($_ = readline $fh) && !/^ob=/;
# put the last line back, we read too far
seek($fh, -length($_), 1);
} else {
print $tmp $_;
}
}
close ($fh);
## Skip all calls to external libraries, as defined by $cob
my $tmpname = $tmp->filename;
open ($tmp, '<', $tmpname) or die $!;
while (readline $tmp) {
if (/$cob/) {
# skip until we find a line starting with calls=
# skip that line too
0 while defined($_ = readline $tmp) && !/$calls/;
# then we skip until we either hit ^word= or an empty line.
# In other words: skip all lines that start with 0x
0 while defined($_ = readline $tmp) && /^0x/;
# put the last line back, we read too far
seek($tmp, -length($_), 1);
} else {
print;
}
}