Re: Avoid spurious failures from <fenv.h> fallback functions

[Andreas Jaeger <aj at suse dot com>]

On 06/20/2013 12:48 PM, Joseph S. Myers wrote:
> Many of the fallback implementations of various <fenv.h> functions in
> math/ indicate failure even when the semantics of the function mean
> there is actually nothing to do in the case where exceptions and
> rounding modes aren't supported.  I've filed bug 15654 for this issue.
>
> This patch changes them to indicate success where semantically
> appropriate.  This is intended to match what the implementations for
> Tile and MicroBlaze do (but covers more functions).  However, the
> fallback functions still use stub_warning, which seems appropriate
> when there isn't substantive support for the underlying
> functionality.  Tested for MIPS64 soft-float (specifically, this fixes
> the stdlib/bug-getcontext failure previously seen in that case).
>
> Kaz, Thomas, note that there is no SH version of fgetexcptflg.c.  This
> looks like an oversight, given that all the other <fenv.h> functions
> have SH implementations.  Neither old nor new stub version would
> really be correct for SH; there should be a real implementation.
> Filed as bug 15655.
>
> Other than for Tile and MicroBlaze (with partial sets of their own
> fallback functions) and SH (as noted above), it appears that these
> fallback functions are only used as a set (all used together or none
> used at all), in cases lacking hardware support for exceptions or
> rounding modes.

I'm fine with this approach,

Andreas
--
 Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi
  SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
    GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126