Hi,

To get better feeling for how well line info is represented in dwarf. I wrote two more die_checks. One to check that DW_AT_decl/call file, line and column come in understandable pairs. That only found issues in some test files, but none in practice with modern gcc. And another that checks that all start addresses of code block dies are also mentioned in the line info table. That one does produce some real warnings. Spot checking reveals these are also the addresses at which gdb gets somewhat confused at which source line the code really is.

I also added a small addition to dwarf::attribute_type. find_integrate is something I keep writing myself each time. I didn't happen to use them in these particular new tests, but I do use it during debugging (for example to get at the name of a die construct). And it seemed useful in general.

Cheers,

Mark

commit 849c6da0eeb1b453c7bb7ffea25a16c78d7e0a1b Author: Mark Wielaard mjw@redhat.com Date: Wed Apr 13 12:33:29 2011 +0200

dwarflint: new check_die_line_info die check.

Check that each code block start address is also mentioned in the line table.

commit d7c3ae057ab527750d172173b8d2a8639be6e823 Author: Mark Wielaard mjw@redhat.com Date: Wed Apr 13 12:26:44 2011 +0200

dwarflint: new check_die_decl_call.cc die_check.

Check that each decl or call attribute come in file/line and column/line pairs.

commit 1f282b3fd8fcde13ba8acf21aca9e40e4328c222 Author: Mark Wielaard mjw@redhat.com Date: Tue Apr 12 15:01:02 2011 +0200

Add elfutils::dwarf::attribute_type::find_integrate ().

Same as find (), but if the attribute name isn't found, but there is is an abstract_origin or specification attribute, then will try to find_integrate () the name from that reference.

check_linkage_external_die.is_external () now uses it.