Bug#281815: glibc: Please change cryptic error message: "Servname not supported for ai_socktype"

[GOTO Masanori <gotom@debian.or.jp>]

At Mon, 29 Nov 2004 15:56:48 -0200,
Henrique de Moraes Holschuh wrote:
> On Tue, 30 Nov 2004, GOTO Masanori wrote:
> > At Wed, 17 Nov 2004 21:20:09 -0200,
> > Henrique de Moraes Holschuh wrote:
> > > It is nearly impossible for an user to understand what the hell "Servname
> > > not supported for ai_socktype" means.  Please at least change that to
> > > "Socket service name unknown", or something else that makes sense to someone
> > > who DOES NOT write Unix socket code...
> > 
> > I don't think it's big difference between the original message and
> > your proposed one.  Note that at least FreeBSD and Solaris8 uses your
> 
> There is a quite big one, Goto-san.  "Servname" is not defined anywhere, it
> does not exist as a word, it is not on the top of a comment in
> "/etc/services", and it is not documented in services(5) either. It is not
> documented even in any of the getserv* manpages.

Well, other Unices define "servname" in their manpages.  However,
glibc does not provide manpages (and manpages is not related with
glibc).  Unfortunatelly libc.info does not provide getaddrinfo(3),
but when we write getaddrinfo entry, we should use "servname".

> "Service name" is at least something that can be understood by someone that
> knows what a "service" is. And the fact that there is a 'services' database
> in nsswitch.conf and there is an /etc/services file, and that apropos
> services gives an useful pointer to services(5) makes a LOT of difference.
> 
> IMHO a very good error message would be something like:
> "Service name %s not in the system socket service name database".  

That makes sense, but I don't judge it's really useful modification or
not..

Regards,
-- gotom