Re: RFC: Testing different long double formats

[Joseph Myers <joseph at codesourcery dot com>]

On Mon, 2 Jul 2018, Florian Weimer wrote:

> The attached patch attempts to cover the -mlong-double-64 case.  Is this a
> direction in which we want to move?

I think we want to move in the direction Tulio gave in 
<https://sourceware.org/ml/libc-alpha/2018-06/msg00686.html>: once all the 
ldouble tests are built for both -mabi=ieeelongdouble and 
-mabi=ibmlongdouble, it should be straightforward to build them for 
-mlong-double-64 as well for ldbl-opt configurations (and since my commit 
bc9620d040b7494f457ccb750c9797b47ed76ada to build them for long double = 
double configurations, they should just work when built with 
-mlong-double-64, including using the correct ulps).

Much the same applies for other -mlong-double-64 redirections: the 
-mabi=ieeelongdouble test infrastructure should be added in a form that 
makes it each to add ldbl-opt testing of the same tests with 
-mlong-double-64.

> The test needs Jakub's patch add:
> 
>   <https://sourceware.org/ml/libc-alpha/2017-08/msg01131.html>

Please commit that patch with such a wrapper for libio/bits/stdio-ldbl.h 
as well.  (It's possible that stdio-common/bits/printf-ldbl.h, 
misc/bits/syslog-ldbl.h, stdlib/bits/monetary-ldbl.h will only end up 
being needed in tests run within those directories, and likewise new 
argp-ldbl.h, err-ldbl.h, error-ldbl.h that don't yet exist, though there 
would be nothing wrong with having wrappers for them as well so that 
linknamespace or check-installed-headers tests could be run for long 
double variants if desired.)

-- 
Joseph S. Myers
joseph@codesourcery.com