Re: [lmi] [lmi-commits] master 98a3720 2/6: Explicitly qualify std::size_t

[Vadim Zeitlin]

On Wed, 17 Feb 2021 21:22:02 +0000 Greg Chicares <gchicares@sbcglobal.net> 
wrote:

GC> On 2/17/21 1:06 PM, Vadim Zeitlin wrote:
[...]
GC> > finds no matches right now, so it looks like a check for this one would be
GC> > pretty simple. This wouldn't detect (even not commented out) occurrences 
of
GC> > size_t at the end of lines, but I don't think there is a real risk of this
GC> > happening in practice.
GC> 
GC> Okay, committed and pushed: 24c32d6ed..8d4034b90  master -> master

 Thanks!

GC> For the record, I tested this alternative:
GC> 
GC> #if 0
GC>     boost::sregex_iterator i(f.data().begin(), f.data().end(), r);
GC>     boost::sregex_iterator const omega;
GC>     for(; i != omega; ++i)
GC>         {
GC>         boost::smatch const& z(*i);
GC>         if
GC>             (f.leaf_name() != "test_coding_rules.cpp"
GC>             )
GC>             {
GC>             std::ostringstream oss;
GC>             oss
GC>                 << "contains unqualified 'size_t': '"
GC>                 << z[0]
GC>                 << "'."
GC>                 ;
GC>             complain(f, oss.str());
GC>             }
GC>         }
GC> #endif // 0
GC> 
GC> but rejected it:
GC>  - it's slower prima facie
GC>  - it gives very little more information
GC>  - we'll seldom be writing 'size_t' anyway

 Agreed on all points.

GC> >  Err, gcc definitely requires __cxa_demangle to have this prototype and on
GC> > the platforms where size_t and int have different size (such as Win64), it
GC> > would be a fatal mistake to use int here. Similarly, we still need to use
GC> > size_t when overriding base class virtual functions using it.
GC> 
GC> The change I propose is not
GC> 
GC> -     ,std::size_t* length        // just use 0
GC> +     ,int*         length        // just use 0

 And I've indeed completely misunderstood you and indeed thought you were
proposing this, thanks for understanding my misunderstanding correctly and
sorry for it in the first place.

GC> but rather
GC> 
GC> -// ABI:
GC> +// For reference, the ABI specifies this prototype:
GC> +// extern "C" char* __cxa_demangle
GC> +//     (char const * mangled_name  // mangled name, NUL-terminated
GC> +//     ,char       * output_buffer // just use 0
GC> +//     ,std::size_t* length        // just use 0
GC> +//     ,int        * status        // zero --> success
GC> +//     );
GC> 
GC> But the change is a little larger than that, and rather than
GC> type it all here, I'll just commit it (master 064c03382).

 This is indeed a good change, thanks,
VZ

pgpWk4kDhNFe_.pgp
Description: PGP signature