17:41 <vibrys> I'm using result of ucallers(10) as key to assiociation array. Then trying to print stack with print_usyms(val), but lines like "0x7fc51028342d : 0x7fc51028342d" are printed. sprint_ubacktrace() does the job on the other hand, but I do not want to use it as key to association map. I'm running systemtap 3.1. Do You have any ideas?

18:05 <fche> hi vibrys

18:05 <fche> does sprint_ubacktrace() go back to 10 levels of callers, and give you symbol names?

18:09 <fche> long shot, but does putting a probe never { sprint_ubacktrace() } into your first program change anything?

18:09 <fche> are you running stap with the same set of -d options (to load symbol/unwind data for the executables / shared libraries of interest) ?

18:34 <vibrys> fche: sprint_ubacktrace() does give me the symbol names. I'm running stap with the same set of -d options. I just compiled systemtap 3.2 and expreriencing the same bevaiour. Now checking { sprint_ubacktrace() } ...

18:35 <fche> does the 'probe never ...' help?

18:35 <vibrys> i'm now checking it ...

18:35 <fche> ok; 'stap -vv ...' or even '-vvv' may help narrow down the cause

18:37 <vibrys> { sprint_ubacktrace() } does not help

18:38 <fche> hm how about a sprint_ubacktrace() from the same place where you are collecting ucallers(10) etc. ?

18:38 <fche> hm, print_usyms ... which context are you calling that from ?

18:39 <fche> if it's from something like a 'probe end', then print_usyms won't have the right context (the current process and its symbol table etc.)

18:39 <fche> the docs do not spell this out though :&

18:40 <vibrys> yes, I'm calling it from probe end with hope to postpone heavy work away from runtime probes

18:40 <fche> aha ok that makes sense then

18:40 <fche> yeah the problem is that the heavy work requires context of the threads that generated those hex addresses

18:40 <fche> otherwise they could've come from any other random process

18:46 <fche> https://sourceware.org/bugzilla/show_bug.cgi?id=10514 <-- this is probably it, very old RFE

18:47 <vibrys> I understand. So I did the following to remember full stack traces within the body of my runtime probe: theTrace = ucallers(10); if (theTrace in stack_traces) next; stack_traces[theTrace] = sprint_ubacktrace()

18:47 <fche> aha good plan

18:49 <vibrys> fche, thank You for Your help and explanation of what's going on here

18:49 <fche> sorry I don't have a better answer :(

18:49 <fche> but if your solution works well enough, you might want to add it as a comment to that PR, so we find it for others

18:50 <vibrys> OK, I'll do so

18:50 <fche> thanks

19:54 <fche> looks good, thanks!