Re: GCC 8.3.0, -flto and violation of C++ One Definition Rule

[Xi Ruoyao <xry111@mengyan1223.wang>]

On Wed, 2021-12-29 at 06:44 -0500, Tom Kacvinsky via Gcc-help wrote:
> Hi,
> 
> First, using GCC 8.3.0 and binutils 2.37.I am trying to increase
> performance of linking our product, so I thought I'd give LTO a try.  So
> I am compiling all object files with -flto, and passing -flto to g++
> (which we use as our link driver).  However, what I have found is that
> some of our code violates the C++ One Definition Rule (-Werror=odr). This
> only happens when building with LTO - without LTO, the C++ rule is
> not violated.

No, it's violated all the time.  -Wodr only works with LTO because it's
impossible to detect ODR volations (across multiple TUs) just by linking
normal (non-LTO) object files.

> The problem exists with LTO using both the BFD and gold
> linkers.
> 
> So, my question is, since the LTO object files are now such that one
> needs to use gcc-nm to examine them (which I know is a wrapper around nm,
> and passes an option to load the LTO plugin). how can I leverage that to
> see if there are other translation units that define the class that ODR
> violation is complaining about?
> I did do a fairly thorough analysis of
> the object files and did not see there the particular class and methods
> would be multiply defined, but that was just based on symbol names from
> gcc-nm output.

Generally you can't find ODR violations this way.  ODR violation does
not mean multiple definitions of a symbol.  For example:

// in 1.cpp

enum A
{
	a = 1,
};

// in 2.cpp

enum A
{
	a = 2,
};

If you link the objects compiled from these TUs together, you will
violate ODR.  But these TUs won't produce any symbols for A or A::a at
all.

> I suspect there is more to this since the object files
> have LTO information now, and that is what I'd like to examine.
> 
> Any hints on how to move forward with diagnosing LTO link errors?

With GCC-11, the diagnostic shows the exact location of ODR violation,
like:

> t-1.cc:1:6: warning: type 'A' violates the C++ One Definition Rule [-Wodr]
>     1 | enum A
>       |      ^
> t-2.cc:3:6: note: an enum with different value name is defined in another translation unit
>     3 | enum A
>       |      ^
> t-1.cc:3:9: note: name 'a' is defined to 1 while another translation unit defines it as 2
>     3 |         a = 1,
>       |         ^
> t-2.cc:5:9: note: mismatching definition
>     5 |         a = 2,
>       |         ^
> 

I'm not sure how GCC 8 behaves.

-- 
Xi Ruoyao <xry111@mengyan1223.wang>
School of Aerospace Science and Technology, Xidian University