[Bug default/28641] unspecified indirect subtype changes.

["woodard at redhat dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28641

--- Comment #4 from Ben Woodard <woodard at redhat dot com> ---
The only meaningful difference that I can see in one case is "const"

excerpt from the ABIXMLs:

GCC 

          <member-function access='public' const='yes' vtable-offset='6'>
            <function-decl name='has_changes'
mangled-name='_ZNK7abigail10comparison10array_diff11has_changesEv'
                filepath='../../../../Work/libabigail/src/abg-comparison.cc'
                line='3578' column='1' visibility='default' binding='global'
size-in-bits='64'
elf-symbol-id='_ZNK7abigail10comparison10array_diff11has_changesEv'>
              <parameter type-id='type-id-1051' is-artificial='yes'/>
              <return type-id='type-id-7'/>
            </function-decl>
          </member-function>

clang:
          <member-function access='public' vtable-offset='6'>
            <function-decl name='has_changes'
mangled-name='_ZNK7abigail10comparison10array_diff11has_changesEv'
               
filepath='/home/ben/Shared/test/libabigail-test/l5/../../../Work/libabigail/include/abg-comparison.h'
                line='1374' column='1' visibility='default' binding='global'
size-in-bits='64'
elf-symbol-id='_ZNK7abigail10comparison10array_diff11has_changesEv'>
              <parameter type-id='type-id-828' is-artificial='yes'/>
              <return type-id='type-id-2'/>
            </function-decl>
          </member-function>

Here is another example:

gcc:
          <member-function access='public' const='yes' vtable-offset='7'>
            <function-decl name='has_local_changes'
mangled-name='_ZNK7abigail10comparison14type_diff_base17has_local_changesEv'
               
filepath='/home/ben/Shared/test/libabigail-test/g5/../../../Work/libabigail/include/abg-comparison.h'
line='1080' column='1' visibility='default' binding='global' size-in-bits='64'>
              <parameter type-id='type-id-1232' is-artificial='yes'/>
              <return type-id='type-id-8859'/>
            </function-decl>
          </member-function>

clang:
          <member-function access='public' vtable-offset='7'>
            <function-decl name='has_local_changes'
mangled-name='_ZNK7abigail10comparison14type_diff_base17has_local_changesEv'
               
filepath='/home/ben/Shared/test/libabigail-test/l5/../../../Work/libabigail/include/abg-comparison.h'
line='1080' column='1' visibility='default' binding='global' size-in-bits='64'>
              <parameter type-id='type-id-1004' is-artificial='yes'/>
              <return type-id='type-id-8200'/>
            </function-decl>
          </member-function>

So arguably this is a clang bug in that it doesn't seem to record the
const-ness of the function in the DWARF (or we have a parsing bug where we
don't pick it up -- haven't checked that yet). 

I believe that we had a discussion a few years ago about whether const-ness of
a function actually amounted to an ABI difference. I can't recall the exact
ultimate decision on that point.

At the very least, when the difference between the two functions is "const"
then libabigail should print that out rather than emitting nothing.

-- 
You are receiving this mail because:
You are on the CC list for the bug.