On 05/19/2017 02:31 AM, Philipp Stephani wrote:
> Apparently on some systems mbstate_t is a nested struct, and the compiler 
> warns about missing braces. Note that memset to initialize a mbstate_t is 
> explicitly recommended in the libc manual:

Yes, of course memset works (which is all that the glibc manual really says). 
It's just that it's verbose and the verbosity isn't needed.

I guess the problem here is that clang's -Wmissing-braces option generates false 
alarms. The Clang folks may fix that someday. In the meantime I installed the 
first attached patch, to turn off that option for Clang.

> you could now add
> eassert (handlerlist == *dummy);
> or so to the cleanup function.

Thanks, good idea, installed in the 2nd attached patch.

> I think none of the four branches there work for macOS. 
> https://developer.apple.com/legacy/library/documentation/Darwin/Reference/ManPages/man2/pathconf.2.html is 
> silent about case sensitivity, so we're already relying on some undocumented 
> functionality. The getattrlist method is at least documented 
> (https://developer.apple.com/legacy/library/documentation/Darwin/Reference/ManPages/man2/getattrlist.2.html), 
> it just needs to be implemented correctly. I'd suggest to only use a working 
> version of getattrlist on macOS.

I installed that in the 3rd attached patch, at least I hope it works. But better 
yet, let's drop the getattrlist stuff entirely, as it doesn't seem to be needed 
and is just wasting our maintenance time. I installed that in the 4th attached 
patch.