Re: killing gb_Library_add_noexception_objects?

On Monday 15 of July 2013, Bjoern Michaelsen wrote:
> Hi,
>
> a quick git grep suggests that we use gb_Library_add_noexception_objects
> only once, thus I dont expect the original reason for keeping support for
> this (possibly better performance and smaller binaries by skipping stack
> unwinding logic) to be manifesting anymore for us.

 At least with GCC there should not be any performance different 
for -fexceptions vs -fno-exceptions code, except for the size of the binary. 
It is a major contributing factor to the binary size though (10-20% IIRC). No 
idea about MSVC.

> As compiling some special-cased C++ objects without exception support,
> while the majority supports exceptions might lead to confusing during
> debugging, should we kill gb_Library_add_noexception_objects completely and
> only support C++ with exceptions. Note that if there really is code that
> has huge wins from not having exception handling, having those in plain old
> C is still possible.

 I seem to remember a discussion where it was said that C building could be 
dropped, since building plain old C files as C++ is possible :).

> So in the interest of the principle of least surprise, can we always have
> exception handling for C++ code? Opinions?

 Given that's it's practically unused, the answer appears to be simple. If 
somebody in the future will be willing to spend the time working on this 
optimization, they can bring it back.

-- 
 Lubos Lunak
 l.lunak@suse.cz