Re: dwarflint semantic DIE checks

[Petr Machata <pmachata at redhat dot com>]

Roland McGrath wrote:
>> I put some time into that over the week.  The code now includes maps of 
>> expected attributes (per DIE tag), and expected forms (per attribute form).
> 
> You should not need the latter, at least not done in that way.  The
> dwarf::location_attr interfaces already encode that knowledge, in
> dwarf::attr_value::what_space (c++/values.cc).  Actually the knowledge is
> in c++/dwarf-knowledge.cc, which you might use directly.  (That file is
> where I intended to have tables of expected children tags, etc. too.)

>> Attribute form is not checked, because there's no way to obtain that 
>> from <dwarf>.  So the table is here, but it's not being used currently.
> 
> Don't check for "expected form".  Check for "expected value space".
> attr->what_space () returns a space or throws an exception for ambiguity.
> (The exception types still need to be worked out, but for now catching any
> exception from what_space () as meaning "unexpected" is fine.)  The decoder
> is liberal in cases without ambiguity, so you still want to check the
> value_space returned against the expected_value_space () mask.
> But you can let it entirely take care of the name+form->value_space mapping.

That's now done, dwarflint checks for value space mismatch.  For every 
CU, I'm now getting this:

warning: .debug_info: DIE 0xb: in "compile_unit" attribute "name" has 
value of unexpected type "identifier".

Looking into values.cc, that's probably related to this:

   unsigned int expected = expected_value_space
     (dwarf_whatattr (thisattr ()), 0); // XXX need tag!

I tried to trick it by using (*reference()).tag() instead of the 0, but 
that just seems to throw an exception "no reference value" 
(DWARF_E_NO_REFERENCE).  I didn't dig in that further.

PM

Attachment: signature.asc
Description: PGP signature