uprobes: make C++ symbol demangling explicit

[viktormalik]

Introduce a new part of uprobe specification which allows to explicitly enable language-specific features. For now, C++ prefixes (:c++ and :cpp) are supported. Specifying one of these turns on symbol demangling which is now disabled by default.

Some examples of the current behaviour

# cat cpptest.cpp
extern "C" int fun() { return 1; }
void fun(int x) {}
void function() {}
class Foo { void fun() {} };
class Bar { void fun() {} };

# nm cpptest
0000000000401126 T fun
0000000000401148 T _Z3funi
0000000000401131 T _Z8functionv
0000000000401184 W _ZN3Bar3funEv
0000000000401178 W _ZN3Foo3funEv

# 
bpftrace -e 'u:./cpptest:fun { exit(); }'
Attaching 1 probe...    (C "fun" only)

bpftrace -e  'u:./cpptest:c++:fun { exit(); }
Attaching 3 probes...   ("fun", "fun(int)", "function")

bpftrace -e 'u:./cpptest:c++"fun(int)" { exit(); }' 
Attaching 1 probe...    ("fun(int)")

bpftrace -e 'u:./cpptest:c++:*fun { exit(); }'
Attaching 5 probes...   (all the functions)

bpftrace -e 'u:./cpptest:c++:"*::fun" { exit(); }'
Attaching 2 probes...   ("Foo::fun" and "Bar::fun")

While this is not perfect (e.g. the second example should not match function), it's better than what it was before (see #2186). I'd prefer using this as a starting point and then following with more PRs rather than doing everything at once.

Missing features include:

• more sophisticated matching of mangled symbols (i.e. fix the above 2nd example)
• better support for namespaces (e.g. support ::fun)
• support for demangling in probe listing

Note that this is a breaking change b/c demangling is not done by default now. However, since it was never documented and it sometimes had incorrect and unexpected behaviour, I believe that this is the correct step forward.

Resolves #2186.

Checklist

• Language changes are updated in man/adoc/bpftrace.adoc and if needed in docs/reference_guide.md
• User-visible and non-trivial changes updated in CHANGELOG.md
• The new behaviour is covered by tests

[mmarchini]

overall LGTM, definitely a breaking change as you said (but that seems fine in this case). Have you looked into how much work it would be to fix the 2nd example? That's the only missing thing I think would be worth including in this PR if the fix is not a large patch.

[viktormalik]

Have you looked into how much work it would be to fix the 2nd example? That's the only missing thing I think would be worth including in this PR if the fix is not a large patch.

For this particular case, it's as simple as adding an extra ( to the end of the query. There are other cases, though, that'll require more thinking. Let me have a look, if we manage to review and merge this in the meantime, I'll send the fix as a follow-up PR ASAP.

[viktormalik]

Have you looked into how much work it would be to fix the 2nd example? That's the only missing thing I think would be worth including in this PR if the fix is not a large patch.

For this particular case, it's as simple as adding an extra ( to the end of the query. There are other cases, though, that'll require more thinking. Let me have a look, if we manage to review and merge this in the meantime, I'll send the fix as a follow-up PR ASAP.

It turned out not to be too complicated to fix the bugs that I know of, so I updated this PR (see 17beb2c).