Re: Poor me... (was: Re: [PATCH, RFA] Reworked fix for PR4402)

[Paolo Carlini <pcarlini at unitus dot it>]

Nathan Myers wrote:

> On Sun, Dec 02, 2001 at 02:00:56AM +0100, Paolo Carlini wrote:
> > Nathan Myers wrote:
> >
> > > On Sun, Dec 02, 2001 at 01:26:40AM +0100, Paolo Carlini wrote:
> > > > of course there is *no* grouping at all for float types...
> > > > (see, f.i., Langer/Kreft 571-572).   May I say that I have
> > > > been a little bit sidetracked by the comment at the beginning of
> > > > _M_widen_float?
> > >
> > > Not so fast.  The restriction of grouping to integer types is
> > > manifestly a typo in the standard, and the subject of a Defect
> > > Report.  Since the standard unambiguously requires parsing
> > > separators in floats, there's little room to argue that the
> > > restriction was intended and understood by the comittee.
> >
> > ... the *present* libstc++-v3 code does *not* group
> > float types and that comment is not coherent with the code. Do you
> > believe we should start implementing that? What's the official status
> > of the Defect Report? It could be an interesting project.
>
> The Standard requires parsing separators in floats, and there is no
> prospect of that changing.  The official status of the Defect Report
> is still "Open"; resolution has been stalled for years by several
> members of the LWG, but they can't put it off forever.  It is obvious
> we must implement parsing of separators in floating point types.  IMHO
> we should also implement formatting with separators for floating point
> types, in anticipation of the expected resolution.

I see.

> My memory says that support was in there before.  If so, somebody must
> have ripped it out without changing the comment.

Interesting... By "before" you mean in the old good times of v2 ?

> > By the way, what do you think of the "exponential allocation" patch?
> > Without your neat statements in July, probably we would'nt be working
> > so hard at it!
>
> I have been following it with great interest, wishing I could afford
> the time to participate more.  Now get some sleep!

Will do :-)
As soon as you find some spare time, please have a look to the current proposal for the
allocator, it seems to me quite satifying and the performance are very good:

    http://gcc.gnu.org/ml/libstdc++/2001-12/msg00007.html

Thanks,
Paolo.



>
>
> Nathan Myers
> ncm at cantrip dot org

--
--------
Paolo Carlini
Universita' degli Studi della Tuscia
Largo dell'Universita', snc - I-01100 Viterbo - Italy
Tel. +39 0761357027 - Fax. +39 0761357179