where do stack traces get filled in?

[Andrew Haley]

Tom Tromey writes:
 > >>>>> "Adam" == Adam Megacz <gcj@lists.megacz.com> writes:
 > 
 > Adam> Are backtraces a target-specific thing (which would have to be
 > Adam> rewritten for win32), or would I be able to do some clever
 > Adam> cut-and-paste out of glibc? Or is this impossible with SJLJ
 > Adam> exceptions?
 > 
 > Adam> Well, I guess there's licensing issues with glibc (LGPL vs
 > Adam> GPL-with-exception); perhaps "review the glibc code, learn how
 > Adam> it works, and then reimplement the same algorithm".
 > 
 > I don't really know what exactly would have to be done.
 > For all I know there is some Win32 API you could use here.

Walking the x86 stack is really pretty easy -- it's just a list of
stack frames.  Other processors can be much harder.

What I would like, in the general case, is to expose the logic in
gcc's exception unwinder so that it can be used by backtrace.

 > Occasionally we've discussed being able to find the backtraces
 > using information that is already around, for instance
 > exception-handling info.

Yup.

 > We need a more powerful approach because we need to be able to find
 > the classes whose methods are on the stack; this is used by some of
 > the Java security features we don't implement.

I've wanted to do this for ages, but I never seem to find the time.  I
might be able to do it over the holidays,

Andrew.