Re: libobjc2 fails to build with gcc 5.4.0

[Niels Grewe]

On 03.04.2017 13:06, David Chisnall wrote:
> On 3 Apr 2017, at 11:51, Ivan Vučica <address@hidden> wrote:
> > 5.4 is almost a year old, so maybe this was fixed. However, I'm not sure it 
> > would be so bad to have a workaround for this compiler?
> >
> > There is already a warning that clang should be used (which I, of course, 
> > switched to; I prefer it anyway). But having a workaround to make it work with 
> > GCC, even in a basic, unsupported way, is probably still good?
>
> As I said in the pull request, I’m hesitant to work around this for three 
> reasons:
>
> - It’s a bug in GCC
> - The work-around code requires using macros and loses type safety
> - A runtime compiled with GCC lacks a number of important features
>
> I think simply removing support for building with GCC is probably a better idea.

As I said in the pull request, I don't have any interest in GCC as an 
Objective-C compiler anymore (hell, I haven't compiled a single line of 
Objective-C code using gcc this year…). But I think outright dropping 
GCC support is a bit harsh. You'd be turning up the heat a lot on 
existing users of libobjc2 in gcc-based environments [0]. It would also 
be a half-baked measure without removal of the legacy runtime API 
(objc_msg_lookup() and friends). My favoured approach would be 
deprecation of the old runtime API (and GCC support) in the next 
release, and removal in the release after that.

Cheers,

Niels

--
[0] Even if you're stuck with manual reference counting etc., libobjc2 
buys you a lot of good improvements, such as thread-safe +initialize…