Front page | perl.perl5.porters |
Postings from December 2012
Re: SvUPGRADE and void
Thread Previous
|
Thread Next
From:
Peter Martini
Date:
December 14, 2012 19:59
Subject:
Re: SvUPGRADE and void
Message ID:
CAFyW6MTSV5O2zKGYcD1-EtmjJBz1xGK=Dx9UdQOPBzaxw_NhBg@mail.gmail.com
On Fri, Dec 14, 2012 at 12:48 PM, Dave Mitchell <davem@iabyn.com> wrote:
> On Fri, Dec 14, 2012 at 03:22:15PM +0000, Dave Mitchell wrote:
>> On Fri, Dec 14, 2012 at 01:22:42PM +0000, Nicholas Clark wrote:
>> > On Thu, Dec 13, 2012 at 01:14:28PM +0000, Dave Mitchell wrote:
>> >
>> > > The first is to stick a (void) in front of every bare SvUPGRADE() in the
>> > > the perl distribution (including dist/ and, via rt.cpan.org, cpan/).
>> > > This seems like quite a bit of effort and is a retrograde step.
>> >
>> > And, arguably, on every distribution on CPAN that also generates a warning?
>> >
>> > > The second is to bite the bullet and change SvUPGRADE from an expression
>> > > into a statement; i.e.
>> > >
>> > > #define SvUPGRADE(sv, mt) (SvTYPE(sv) >= (mt) || (sv_upgrade(sv, mt), 1))
>> > >
>> > > becomes
>> > >
>> > > #define SvUPGRADE(sv, mt) if (sv->type >= (mt)) { sv_upgrade(sv, mt); }
>> > >
>> > > A search for 'if.*SvUPGRADE' on grep.cpan.me indicates that about 15
>> > > distributions would fail to compile and need fixing.
>> > >
>> > > I'm open to other suggestions.
>> >
>> > I think we should bit the bullet and go for it. For anything that needs
>> > changing, it's a build-time error from the C compiler? (with an easy fix)
>> > That's not going to break anything in production.*
>>
>> Ok, I've bitten the bullet.
>
> ... and broken g++.
>
> Can anyone suggest a definition for SvUPGRADE that works for both of the
> forms
>
> SvUPGRADE(...);
> (void)SvUPGRADE(...);
>
> on gcc, clang and g++, and doesn't generate warnings?
>
> My new incarnation,
>
> #define SvUPGRADE(sv, mt) \
> STMT_START { if (SvTYPE(sv) < (mt)) sv_upgrade(sv, mt); } STMT_END
>
> which expands this this line in regcomp.c:
>
> (void)SvUPGRADE(sv_dat,SVt_PVNV);
>
> into:
>
> (void)do { if (((svtype)((sv_dat)->sv_flags & 0xff)) < (SVt_PVNV)) Perl_sv_upgrade(my_perl, sv_dat,SVt_PVNV); } while (0);
>
> gives this error under g++:
>
> regcomp.c:8653:35: error: expected primary-expression before ‘do’
> regcomp.c:8653:35: error: expected ‘;’ before ‘do’
>
> while the old version:
>
> #define SvUPGRADE(sv, mt) (SvTYPE(sv) >= (mt) || (sv_upgrade(sv, mt), 1))
>
> gave warnings under clang for this form:
>
> SvUPGRADE(...);
I can't think of a suitable no-op, but let's say we have a function:
void do_nothing(void){;}
Couldn't we make SvUPGRADE into:
do_nothing; STMT_START { if (SvTYPE(sv) < (mt)) sv_upgrade(sv, mt); } STMT_END
?
That should satisfy the sanity checks of both, and I would think the
do_nothing should be optimized away.
>
>
> --
> Diplomacy is telling someone to go to hell in such a way that they'll
> look forward to the trip
Thread Previous
|
Thread Next