CGAL users discussion list

[Marc Glisse <>]

On Sat, 7 Dec 2019, Fritz Mueller wrote:

> > > > Looking into the sources a bit, get_static_error_handler() etc. are define in 
> > > > header-only mode as inlines, returning references to local statics.  I think 
> > > > this means each call could conceivably use different storage for those 
> > > > statics,
> > >
> > > No. A static variable inside a function is a unique global object. It is the 
> > > usual way to simulate C++17 inline variables and get singletons.
> >
> > Okay, good to know this is true, even for inlines -- thanks.  I can do some 
> > experiments with a small bit of test code to verify that I understand 
> > correctly.
>
> Well, what I’ve found with OS X clang so far is this:
>
> Static variables inside a function are indeed unique if that function is 
> inline, but seeming *not* if that function is static inline?  The CGAL 
> functions in question are perhaps effectively static inline because the 
> enclosing anonymous namespace.

Uh, what is that anonymous namespace doing there? It defeats the purpose 
of the inline functions. It seems that the namespace was there first, used 
to hide some private stuff but not necessary, and people didn't notice it 
when they added the header-only mode.

Thanks a lot for the detailed analysis!

If you want to make sure we don't forget to fix it, you can open an issue 
in cgal's github with this information (or even better open a pull request 
with a fix), although hopefully it will get fixed soon even without that.

--
Marc Glisse