Re: [PATCH 2/2] Add C++ iterators

[Petr Machata <pmachata at redhat dot com>]

Mark Wielaard <mjw@redhat.com> writes:

> Is that a common use case? I would have imagined that you would iterate
> over just the Dwarf_Dies of one particular unit separately or given a

That would be along these lines:

  for (die_tree_iterator it (uit), end (++uit); it != end; ++it)
    ;

> particular Dwarf_Die would like it to iterate over just the die_tree
> under that DIE. Given the current design of the die_tree_iterator how
> would one do that?

This is not implemented, though I admit it would be a useful feature.
At this point however I would very much like to keep the scope of the
patch as it is now.

> Also I think that we really should provide an easy way to walk the
> logical DIE tree. Which means resolving DW_TAG_imported_unit DIEs and
> just continue with the children of the imported unit at that point.
> There should of course also be an option to walk the "raw" DIE tree. But
> it feels wrong to let the user do logical walks on their own, checking
> for imported_unit tags, walking the imported_unit children tree, keep
> their own parent stack, etc.

I agree, this would be tremendously useful--we discussed this recently
on this list after all.  But it should be implemented in C libdw, and
only then (optionally) reused in the C++ wrappers.

> That of course does mean the current internal tracking using just a
> Dwarf_Off is not enough, because resolving imported_units might cross
> Dwarfs.

One option would be to build the logical iterator on top of the raw
iterator, and track import history in a vector of raw iterators.  That's
essentially what dwgrep does.

Thanks,
Petr