Re: usability issues with inlining and backtraces

[Milian Wolff <milian.wolff@kdab.com>]

[-- Attachment #1: Type: text/plain, Size: 2513 bytes --]

On Monday, May 23, 2016 11:11:04 AM CEST Arnaldo Carvalho de Melo wrote:
> Em Mon, May 23, 2016 at 11:10:00AM +0200, Milian Wolff escreveu:
> > Hey all,
> > 
> > I've recently encountered some nasty issues with perf which should - I
> > hope - be possible to be fixed.
> 
> I've quickly read this, all interesting, but I need some time to digest
> it fully, which I don't have right now, I'll get to it as soon as I can.
> 
> The fact that other open source tools do it as you expect is a precious
> observation, as we can go look at how they solved problems.

I have now found something very interesting in the perf code base:

The addr2line implementation with HAVE_LIBBFD_SUPPORT is actually capable of 
producing backtraces that include inliners, similar to what GDB can do. If I 
add debug output to the bfd_find_inliner_info loop inside addr2line in 
srcline.c, and force unwind_inlines to true, then I see e.g.:

~~~~~~~~~~
inline: 0 _ZNSt8__detail4_ModImLm2147483647ELm16807ELm0ELb1ELb1EE6__calcEm (/
usr/include/c++/6.1.1/bits/random.h:143)
inline: 1 _ZNSt8__detail5__modImLm2147483647ELm16807ELm0EEET_S1_ (/usr/
include/c++/6.1.1/bits/random.h:151)
inline: 2 _ZNSt26linear_congruential_engineImLm16807ELm0ELm2147483647EEclEv (/
usr/include/c++/6.1.1/bits/random.h:332)
inline: 3 
_ZSt18generate_canonicalIdLm53ESt26linear_congruential_engineImLm16807ELm0ELm2147483647EEET_RT1_ 
(/usr/include/c++/6.1.1/bits/random.tcc:3332)
inline: 4 
_ZNSt8__detail8_AdaptorISt26linear_congruential_engineImLm16807ELm0ELm2147483647EEdEclEv 
(/usr/include/c++/6.1.1/bits/random.h:185)
inline: 5 
_ZNSt25uniform_real_distributionIdEclISt26linear_congruential_engineImLm16807ELm0ELm2147483647EEEEdRT_RKNS0_10param_typeE 
(/usr/include/c++/6.1.1/bits/random.h:1818)
inline: 6 
_ZNSt25uniform_real_distributionIdEclISt26linear_congruential_engineImLm16807ELm0ELm2147483647EEEEdRT_ 
(/usr/include/c++/6.1.1/bits/random.h:1809)
inline: 7 main (/path/to/test.cpp:14)
~~~~~~~~~~

I would really like to see this information be included in the call graph 
shown by perf, e.g. in the stdio/browser output of perf report, but also in 
perf script etc. pp.

Can someone suggest a way to do that? These entries cannot be represented by a 
normal instruction pointer address, so extending the existing callchains with 
additional entries is not straight forward to do.

Thanks

-- 
Milian Wolff | milian.wolff@kdab.com | Software Engineer
KDAB (Deutschland) GmbH&Co KG, a KDAB Group company
Tel: +49-30-521325470
KDAB - The Qt Experts

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 5903 bytes --]