Re: [PATCH] Fix gnu.ver, fix std::tr1::hash<long double> on {ppc,s390,sparc,alpha}*-linux

[Benjamin Kosnik <bkoz at redhat dot com>]

> The following updated patch seems to cure it.
> Regtested on ppc-linux, additionally tested with
> #include <tr1/functional>
> #include <iostream>
> 
> int
> main (void)
> {
>   std::tr1::hash <long double> h;
>   size_t a = size_t (h (3.141592653L));
>   size_t b = size_t (h (3.141592653E31L));
>   std::cout << a << " " << b << std::endl;
> }
> compiled both using default options and with -mlong-double-64,
> executed against 4.3 libstdc++ with this patch as well as vanilla 4.3
> libstdc++ from gcc build configured --without-long-double-128
> (obviously the -mlong-double-64 compiled program only).
> 
> Ok for 4.3?
> 
> For the trunk, different changes are needed:
> 1) move _ZNKSt4hash* exports to GLIBCXX_3.4.11, similarly
> system_error etc. 2) _ZNKSt4hashI[eg]EclE[eg] needs to be handled
> similarly as _ZNKSt3tr14hashI[eg]EclE[eg] in this patch, though
> perhaps to avoid introducing compatibility-ldbl-0x.cc we could just
> use an alias as _ZNKSt3tr14hashIeEclEe and _ZNKSt4hashIeEclEe are
> identical In any case, _ZNKSt4hashIgEclEg should be exported as
>    GLIBCXX_LDBL_3.4.11.
> 
> BTW, do the changes from yesterday mean that -std=gnu++0x compiled
> #include <functional>
> #include <iostream>
> 
> int
> main (void)
> {
>   std::hash <long double> h;
>   size_t a = size_t (h (3.141592653L));
>   size_t b = size_t (h (3.141592653E31L));
>   std::cout << a << " " << b << std::endl;
> }
> will not link?

Yes.

Here is here is something that just takes care of this and the other
issue you mentioned with the hash<error_code> export. 

Then the system_error stuff will be correct.

It's my suggestion to put this in and then deal clearly with the long
double issue. Sound like a plan?

-benjamin

Attachment: p.20080229-1
Description: Text document