RE: alphasort on HP-NonStop

[Joachim Schmitz]

Thanks, I'll mention this to them.

Bye, Jojo

-----Original Message-----
From: Paul Eggert [mailto:address@hidden 
Sent: Tuesday, December 28, 2010 5:44 PM
To: Joachim Schmitz
Cc: 'Bruno Haible'; address@hidden
Subject: Re: alphasort on HP-NonStop

On 12/28/2010 01:59 AM, Joachim Schmitz wrote:
> I see that gnulib uses strcoll() rather than strcmp(), which seems to be what 
> POSIX mandates.
> I'll request a bug fix for these issues from  HP NonStop development...

When you do that, you might tell them that there's a possible time bomb there.  
(I presume that NonStop developers prefer to avoid time bombs. :-) If strcoll 
fails and set errno == EINVAL, then POSIX allows alphasort to return any value 
it likes, which means that if NonStop's qsort requires a total ordering, their 
scandir cannot call qsort.

They can avoid the problem either by having a strcoll that never fails, or by 
having a qsort that does not require a total ordering, or by having a scandir 
that does not invoke qsort.