[Bug default/30048] incorrect qualifiers within compound types

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

https://sourceware.org/bugzilla/show_bug.cgi?id=30048

Ben Woodard <woodard at redhat dot com> changed:

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

--- Comment #2 from Ben Woodard <woodard at redhat dot com> ---
I would call this one of many DWARF idiom issues. At its base libabigail uses
string comparisons in many cases rather than abstractions of the semantic
information. This can cause problems like this when the expression in the DWARF
is derived the source code and two literally different strings represent
semantically identical items.

I have filed several of these but we haven't gotten to looking at them yet.
Here are some examples:
   https://sourceware.org/bugzilla/show_bug.cgi?id=28641
   https://sourceware.org/bugzilla/show_bug.cgi?id=28664
   https://sourceware.org/bugzilla/show_bug.cgi?id=28675

I find a lot of things like this when trying to test compatibility between
things which are built with different compilers. 

Something like:
abidiff lib_built_with_gcc.so same_lib_built_with_clang.so

The last time that I talked to Dodji about these he suggested either fix the
string comparison or putting in special case code that handles these cases
after the string comparison fails.

The tricky thing is really knowing what are actually ABI artifacts and what are
not. The order of restrict in relation to const doesn't matter but is the fact
that restrict included in the declaration actually an ABI artifact? It wouldn't
affect the calling convention but the codegen could be different if restrict
were not included in the decl. So we don't want to just ignore restrict. What
if a library added 'restrict' to an exported function's parameter. Previous
users of the library may not have been expecting that requirement. Is that
actually an ABI change?

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