Re: 1.10 task: error leaks

Julian Foad wrote on Mon, 11 Dec 2017 19:51 +0000:
> Daniel Shahaf wrote:
> > I think we should enable this warning by default in maintainer mode in
> > supporting compilers, but that's a separate discussion to just fixing
> > these error leaks in trunk (& 1.10.x).
>
> +1. I was going to say that, as soon as I saw this long list.
>
> Anyone want to make it happen?

We already have an instance of __attribute__((warn_unused_result)) in
mod_dav_svn, not guarded by any #ifdef or anything, so I think we don't need to
bother with compiler checks (AC_TRY_COMPILE() and friends), but rather, we can
simply add that attribute unconditionally.

Of course, then we'd have a new problem: we might declare or define a function
without the __attribute__ annotation. It would be best if we could tell our C
compilers, "_Any_ function that returns svn_error_t* should be treated as
having the __attribute__((warn_unused_result)) annotation". Is that possible?

> (I volunteer to do some of this too. In the role of Release Manager I
> need to be asking for volunteers for as much as possible, so please
> don't anyone take offence at my continuing to do so.)

None taken; it should be a team effort. Thanks for leading it.

Cheers,

Daniel
Received on 2017-12-11 21:36:16 CET