Re: Soalris 10 10/2008 SPARC changes

[Martin Sebor <ms...@gmail.com>]

Stefan Teleman wrote:
> On Fri, Aug 14, 2009 at 16:24, Martin Sebor<ms...@gmail.com> wrote:
>> Thanks for the heads up and the patches! I'll review apply them
>> if possible/necessary before the release.
>>
>> I took a quick glance at a few of the diff files and I wonder if
>> you could help me better understand some of the changes in case
>> they could be applied unconditionally.
>>
>> For instance, in ctype.cpp.38.diff, it seems as though the second
>> hunk would be a good change to make regardless (when long long is
>> available). Ditto for the third hunk in locale_body.37.diff (minus
>> the pragmas, of course), and similarly in locale_classic.40.diff
>> and messages.cpp.41.diff.
>>
>> The approach I'm thinking of using is the one you applied in
>> use_facet.h, i.e., defining, say, _RWSTD_ALIGN_MAX_T to unsigned
>> long long, and using it in all the aligned buffers, along with
>> unsigned char for the data.
>>
>> Other than these, can you also help me understand the changes in
>> messages.cpp.41.diff (starting with the third hunk on line 92)?
> 
> That's my bad mistake -- i created a diff for src/messages.cpp from an
> older version of messages.cpp (hacked while i was trying to figure out
> why it was misbehaving so badly on 32-bit SPARCv8).
> 
> I updated the tarball and the messages.cpp.41.diff itself with a valid
> patch now. Sorry for the noise.

No problem. I'll take a look.

> 
> About
> 
> int ret = catclose (pcat_data->catd);
> 
> // ...
> 
> cat_closed = ret == 0;
> 
> My understanding of 22.2.7.1.2, p5 (catalog must be valid) is that  --
> if catclose(3C) fails, it means the catalog was not valid -- therefore
> the C++ library must throw. So, cat_closed is now true if and only if
> catclose(3C) succeeded.  Or is this too restrictive ?

You're right that we should check the value returned from catclose().
I think the facet guarantees that the catalog descriptor is valid so
I think we're safe in that regard. The case where I think there is
a problem is when catclose() fails because of a signal (with EINTR).
That would make __rw_catclose() and the facet dtor leak the catd.
Let me see about fixing that. Thanks for pointing it out!


> 
> About the other patches -- and whether or not they should be applied
> for other platforms besides Solaris/SPARC: i tried not to disturb any
> other ISA's/compiler combinations. If you'd like i can re-write the
> patches for 4.2.2 using the more generic approach you described.

Understood.

Thanks again!
Martin

> 
> --Stefan
>