Re: dwarf and clones

[David B Anderson <davea at quasar dot engr dot sgi dot com>]

>How is this different from an out-of-line instance?  

Not much different conceptually.
We don't have an abstract root, we simply generate dwarf
for the clone as a complete routine,
but the out-of-line instance approach should work ok.

For real inlines and concrete out-of-line instances we follow
the dwarf document.

>It looks like DW_AT_MIPS_clone_origin is the same as DW_AT_abstract_origin.

Yes.

>What would be the effect of describing a clone as a concrete out-of-line
>instance?  

It would not be different, really.
I think we did it differently just because it was thought
to be slighly simpler to implement.
I found some debugger-internal documentation supporting
this recollection.

A key attribute for clones is the nocall attribute.
Such routines may not follow ABI argument-passing rules and
they are not intended for interactive calls from a
debugger in any case.

>DW_AT_MIPS_abstract_name looks like it could be a DW_AT_abstract_origin
>with DW_FORM_ref_addr.  This doesn't seem directly related to clones.

Correct. 


davea@sgi.com