sigaltstack?

[bakinat]

This is great work.
I see you use SA_ONSTACK so I assume you intended to provide an alternate stack (which is better for this case) but I don't see anywhere an initialization with sigaltstack - or maybe I missed something?
Per sigaction man page, in such case the default stack is used (which I assume is as if SA_ONSTACK has not been given)

[vmarkovtsev]

Thanks, sigaltstack() was overseen. Looks likr backtrace() still works on an alternative stack, so I should just add this function call with some allocated mem.

[bakinat]

I don't see how can backtrace() use an alternative stack if sigaltstack wasn't called, but I may be wrong.
Anyway, one more question and few suggestions:

1. Question: How do you determine the size of kNeededMemory?
2. Suggestion: it would be useful to enable the actual printing through some callback (which may default to stderr as you did). For example, if someone would like to write the stack trace into a log file instead.
3. Suggestion: if addr2line fails you can still print a line similar to http://stackoverflow.com/a/1925461
4. Suggestion: you might want to have user control over whether you fork in the handler.
   Thanks again.

[vmarkovtsev]

1. I calculated the worst case memory consumption with the longest lines and added some reserve.
2. Implemented.
3. I will try to research on this, thanks. Looks like some time is needed.
4. Skipping fork() means much instability with multithreaded apps. If can be coded for sure, but I have no enthusiasm with this. A PR would be accepted though.

And the bad news is, sigaltstack() screws backtrace() with glibc, after all. I left the support, but it is disabled by default, since nobody usually wants an empty trace. I'd like to know if there is a way to run backtrace() on a different stack...

[bakinat]

Thanks. I actually agree with you on 4.
About sigaltstack, first I would use SIGSTKSZ and not SIGMINSTKSZ, but I realize this will not help backtrace. Still, in this mode the handler can just print the failure point (using addr2line on siginfo_t->si_addr) with the advantage of handling stack overflow in case of infinite recursion as well.
To make backtrace work, I assume there might be a way to get the pointer to the original stack (ucontext_t->uc_stack?), but it will probably require some work.
Anyway, this looks good enough for now :)
Thanks again!