Re: The patch of glibc which allows the user to override the pfinet server

[Samuel Thibault]

zhengda, le Fri 15 Aug 2008 20:05:53 +0200, a écrit :
> Samuel Thibault wrote:
> >zhengda, le Fri 15 Aug 2008 19:05:49 +0200, a écrit :
> >  
> >>Samuel Thibault wrote:
> >>    
> >>>zhengda, le Fri 15 Aug 2008 18:27:00 +0200, a écrit :
> >>> 
> >>>      
> >>>>+      if (__asprintf (&name, "SOCK_SERV_%d", domain) < 0)
> >>>>+    __libc_fatal ("hurd: Can't get the socket server path\n");
> >>>>   
> >>>>        
> >>>Shouldn't we just return ENOMEM?
> >>> 
> >>>      
> >>In many parts of glibc, __libc_fatal() or assert() is used.
> >>    
> >
> >When there is no way to report the error, isn't it?
> >  
> In the place where __libc_fatal() is called, you're right. They are 
> usually the initialization routines.
> But for the case of assert(), I don't think so. make_hol() is an 
> example.

Because its caller, _help, doesn't have a way to return an error.
And anyway, it's not because some code may be bad that you should do the
same :)

Samuel