Re: Getting inline function information

[Andrew Cagney <cagney at redhat dot com>]

Adam,

Yes, dwarf has two redundant sources of information here.  To be honest 
I think both should be programmably accessable - scanning .debug_line is 
a lot cheaper than .debug_info and often simple file/line info is all 
the user is after.  For instance, fstack would, by default, just want 
.debug_line for backtraces as that keeps things very fast.

One got-ya with .debug_line though, I'm not sure if the spec clearly 
defines inline nesting rules?

Andrew

Adam Jocksch wrote:
> After conversing with Roland this afternoon, I think I've figured out 
> the way we should be approaching this problem. It looks like we need 
> to do the following things for each PC in the stack frame returned 
> from libunwind:
>
> 1) get the scopes for that pc using dwarf_getscopes (already wrapped).
> 2) Started at the innermost scope, work our way backwards and check 
> for a Dwarf_Die with the DW_TAG_inlined_subroutine tag on it. 
> Attributes attached to this tag may give us more information about the 
> call state of the inlined code. You can get the Dwarf tag by using 
> dwarf_tag.
>
> He also suggested taking a look at 
> elfutils/src/addr2line.c:print_dwarf_function (which I have yet to do 
> but will do shortly) for more information. It looks like the things 
> that need to be accomplished to get this working are
> a) Wrap dwarf_tag
> b) Wrap whatever we need to get any/all relevant attributes out of the 
> dwarf_tag (this may very well end up becoming a part of step (a) ).
>
> Hopefully I should be able to get started on this sometime this week, 
> so if anyone has any questions/comments/criticisms please feel free to 
> respond :)
>
> Adam