[Bug default/21296] abidiff reports possibly bogus differences and crashes

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

https://sourceware.org/bugzilla/show_bug.cgi?id=21296

Ben Woodard <woodard at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |woodard at redhat dot com

--- Comment #1 from Ben Woodard <woodard at redhat dot com> ---
jwakely and I were discussing this, this morning.

We think that there are actually three problems:
1) the abidiff crash
2) the second changed function seems to only be different with regards to
whitespace. We think that whitespace differences should be ignored. Can you
make the evaluative function ignore the change in the whitespace.
3) libabigail seems to miss the true nature of the ABI change which is that one
of the parameters is passed on the stack while the other is passed on a
register. We believe that GCC implements the correct calling convention
according to the Itanium C++ ABI.
See: https://bugs.llvm.org//show_bug.cgi?id=23034 I haven't looked at the DWARF
yet but jwakely seemed to say that he wasn't surprised that libabigail missed
detecting this ABI change because the DWARF was insufficient. If that is in
fact the case, then we probably need to file a couple of BZ's to improve the
DWARF on GCC and CLAMG so that libabigail has the information needed to address
this.

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