Re: [patches] Fix *asprintf() to be compatible with *BSD and sanity

["Joseph S. Myers" <joseph@xxxxxxxxxxxxxxxx>]

On Thu, 7 May 2009, Mark Mitchell wrote:

> Joseph S. Myers wrote:
> 
> >>  Here is a, hopefully self explanatory, patch to fix *asprintf() to be
> >> compatible with *BSD and what any sane C programmers expects. Drepper
> >> NAKd it years ago...
> 
> > So I'd suggest raising the issue with WG14, and unless they come up with 
> > good reasons the BSD semantics are problematic then I'm inclined to put 
> > this in EGLIBC as being a compatibility bug fix.
> 
> The WG14 process is not fast.  It might take months to get any kind of
> opinion.

If people are going to reply to a message on the reflector, it does not 
take months for them to do so.

> To me, this change looks like pretty harmless -- it will add a few bytes
> to the library, but no instructions on the critical path (assuming that
> error returns are rare), and could help to avoid Heisenbugs.  There's no
> API/ABI impact, so it seems like this is just a QoI issue.
> 
> What's the harm in putting it in EGLIBC without WG14 sign-off?

The suggestion is not sign-off, it is assessing the views of WG14 
reflector members.  The DTR is silent on whether any change, and if so 
what change, is made to *ptr in the case of error, and cases where 
implementations differ are exactly the sort of case where WG14 can 
usefully discuss which implementation semantics are best to standardise; 
if they then decide to place no requirements in this area, storing NULL 
would seem best, but they could also e.g. distinguish allocation failures 
and other error conditions.

-- 
Joseph S. Myers
joseph@xxxxxxxxxxxxxxxx