cvs emacs and (system-name)

cvs emacs and (system-name)

[Tom Tromey]

I updated Emacs from CVS and built it on my x86 FC6 box.
I ran "emacs -nw -q".
Then in *scratch* I tried (system-name) and got:

(system-name)
"localhost.localdomain"

However, 'hostname' on this machine reports 'opsy'.
So, I suspect this is a bug in Emacs.

FWIW I see this same behavior, on both my machines, with the
Fedora-supplied Emacs 21 builds.

Tom

Re: cvs emacs and (system-name)

[Tom Tromey]

>>>>> "Andreas" == Andreas Schwab <schwab@suse.de> writes:

>> However, 'hostname' on this machine reports 'opsy'.

Andreas> What does hostname -f give you?

opsy. hostname -f
opsy

Tom

Re: cvs emacs and (system-name)

[Andreas Schwab]

Tom Tromey <tromey@redhat.com> writes:

> Then in *scratch* I tried (system-name) and got:
>
> (system-name)
> "localhost.localdomain"
>
> However, 'hostname' on this machine reports 'opsy'.

What does hostname -f give you?

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: cvs emacs and (system-name)

[Tom Tromey]

>>>>> "Chong" == Chong Yidong <cyd@stupidchicken.com> writes:

Chong> Could you try to compile the following program and report its output?

Sigh.  It does report localhost.localdomain.
So, I suppose the bug is my /etc/hosts entry, which I only now
noticed:

127.0.0.1               opsy localhost.localdomain localhost

Thanks, and sorry for the noise.

Tom

Re: cvs emacs and (system-name)

[Chong Yidong]

Tom Tromey <tromey@redhat.com> writes:

>>>>>> "Andreas" == Andreas Schwab <schwab@suse.de> writes:
>
>>> However, 'hostname' on this machine reports 'opsy'.
>
> Andreas> What does hostname -f give you?
>
> opsy. hostname -f
> opsy

Could you try to compile the following program and report its output?


#include <stdio.h>
#include <alloca.h>
#include <string.h>
#include <sys/socket.h>
#include <netdb.h>

int main ()
{
  int hostname_size = 256;
  char *hostname = (char *) alloca (hostname_size);

  while (1)
    {
      gethostname (hostname, hostname_size - 1);
      hostname[hostname_size - 1] = '\0';
      if (strlen (hostname) < hostname_size - 1)
	break;
      hostname_size <<= 1;
      hostname = (char *) alloca (hostname_size);
    }

  if (! index (hostname, '.'))
    {
      struct hostent *hp;
      hp = gethostbyname (hostname);
      if (hp)
	{
	  char *fqdn = (char *) hp->h_name;
	  if (!index (fqdn, '.'))
	    {
	      char **alias = hp->h_aliases;
	      while (*alias && !index (*alias, '.'))
		alias++;
	      if (*alias)
		fqdn = *alias;
	    }
	  hostname = fqdn;
	}
    }
  printf ("%s\n", hostname);
}

Re: cvs emacs and (system-name)

[Richard Stallman]

    Sigh.  It does report localhost.localdomain.
    So, I suppose the bug is my /etc/hosts entry, which I only now
    noticed:

    127.0.0.1               opsy localhost.localdomain localhost


Would someone please put something in etc/PROBLEMS
about this?

Also, maybe init_editfns could detect this and warn the user.

Re: cvs emacs and (system-name)

[Glenn Morris]

Richard Stallman wrote:

> Would someone please put something in etc/PROBLEMS about this?

Doesn't the existing entry cover this:

    *** Emacs does not know your host's fully-qualified domain name.

    You need to configure your machine with a fully qualified domain
    name, either in /etc/hosts, /etc/hostname, the NIS, or wherever
    your system calls for specifying this.

Re: cvs emacs and (system-name)

[Tom Tromey]

>>>>> "GM" == Glenn Morris <rgm@gnu.org> writes:

GM> Doesn't the existing entry cover this:
GM>     *** Emacs does not know your host's fully-qualified domain name.
GM>     You need to configure your machine with a fully qualified domain
GM>     name, either in /etc/hosts, /etc/hostname, the NIS, or wherever
GM>     your system calls for specifying this.

FWIW it isn't entirely clear to me that my config is buggy.
Unfortunately I don't recall how I got it this way -- I suspect I did
a standard Fedora install but did not specify a domain name (IOW, this
may not affect just me).  I don't really have a domain I would put my
machine in, and I think a reasonable number of people may fall into
this category.  Emacs picking "localhost.localdomain" over the actual
name of my machine still seems a bit odd.

Tom

Re: cvs emacs and (system-name)

[Glenn Morris]

Tom Tromey wrote:

> I don't really have a domain I would put my machine in, and I think
> a reasonable number of people may fall into this category. Emacs
> picking "localhost.localdomain" over the actual name of my machine
> still seems a bit odd.

At the moment, we prefer any name with a "." in. We could skip
"localhost.localdomain":

*** sysdep.c	21 Jan 2007 04:18:15 -0000	1.278
--- sysdep.c	19 Apr 2007 05:00:41 -0000
***************
*** 2433,2439 ****
  		/* We still don't have a fully qualified domain name.
  		   Try to find one in the list of alternate names */
  		char **alias = hp->h_aliases;
! 		while (*alias && !index (*alias, '.'))
  		  alias++;
  		if (*alias)
  		  fqdn = *alias;
--- 2433,2440 ----
  		/* We still don't have a fully qualified domain name.
  		   Try to find one in the list of alternate names */
  		char **alias = hp->h_aliases;
! 		while (*alias && ( !index (*alias, '.') ||
! 		     !strcmp (*alias, "localhost.localdomain" ) ) )
  		  alias++;
  		if (*alias)
  		  fqdn = *alias;

Re: cvs emacs and (system-name)

[Richard Stallman]

    > Would someone please put something in etc/PROBLEMS about this?

    Doesn't the existing entry cover this:

	*** Emacs does not know your host's fully-qualified domain name.

	You need to configure your machine with a fully qualified domain
	name, either in /etc/hosts, /etc/hostname, the NIS, or wherever
	your system calls for specifying this.

It seems to be about this situation, but it may be hard to recognize
that this relates.  It would be easier to recognize that if it stated
more concretely what the problem is -- that (system-name) returns
something like "localhost".

Re: cvs emacs and (system-name)

[Stefan Monnier]

>> I don't really have a domain I would put my machine in, and I think
>> a reasonable number of people may fall into this category.  Emacs
>> picking "localhost.localdomain" over the actual name of my machine
>> still seems a bit odd.

But it's kind of difficult for Emacs to guess which of the three names you
put there is "my machine name".  Maybe "localhost" could be ruled out
because it's sufficiently standard.  But "localhost.localdomain" is not
particularly standard, AFAIK.

Maybe the answer is that Emacs should just disregard the "127.*.*.*"
IP range.

> ! 		while (*alias && ( !index (*alias, '.') ||
> ! 		     !strcmp (*alias, "localhost.localdomain" ) ) )

Yuck!!


        Stefan

Re: cvs emacs and (system-name)

[Glenn Morris]

Richard Stallman wrote:

> It seems to be about this situation, but it may be hard to recognize
> that this relates.  It would be easier to recognize that if it stated
> more concretely what the problem is -- that (system-name) returns
> something like "localhost".

done

Re: cvs emacs and (system-name)

[Tom Tromey]

>>>>> "Stefan" == Stefan Monnier <monnier@iro.umontreal.ca> writes:

>>> I don't really have a domain I would put my machine in, and I think
>>> a reasonable number of people may fall into this category.  Emacs
>>> picking "localhost.localdomain" over the actual name of my machine
>>> still seems a bit odd.

Stefan> But it's kind of difficult for Emacs to guess which of the
Stefan> three names you put there is "my machine name".

My recollection is that traditionally the first host name is
considered the canonical one.  'man 5 hosts' backs me up on this :)

Stefan> Maybe "localhost" could be ruled out
Stefan> because it's sufficiently standard.  But "localhost.localdomain" is not
Stefan> particularly standard, AFAIK.

I don't recall seeing 'localhost.localdomain' back in the day -- but I
know *I* didn't put it there.  If Fedora does it then that is standard
enough for me; we're talking about thousands or millions of hosts
here.

Stefan> Maybe the answer is that Emacs should just disregard the
Stefan> "127.*.*.*" IP range.

I don't have any other entry in /etc/hosts.  My machine has no fixed
address and is not in a domain.  This, I believe, is extremely common
among laptop users.

Perhaps if only "localhost" and "localhost.localdomain" are found,
Emacs could fall back to gethostname(2) or uname(2).  With glibc the
latter appears to try to return the domain name as well.

Tom

Re: cvs emacs and (system-name)

[Chong Yidong]

Tom Tromey <tromey@redhat.com> writes:

> Stefan> Maybe "localhost" could be ruled out because it's
> Stefan> sufficiently standard.  But "localhost.localdomain" is not
> Stefan> particularly standard, AFAIK.
>
> I don't recall seeing 'localhost.localdomain' back in the day -- but
> I know *I* didn't put it there.  If Fedora does it then that is
> standard enough for me; we're talking about thousands or millions of
> hosts here.

FWIW, my Debian or Ubuntu boxes don't seem to have this problem.

Re: cvs emacs and (system-name)

[Stefan Monnier]

> ! 		while (*alias && ( !index (*alias, '.') ||
> ! 		     !strcmp (*alias, "localhost.localdomain" ) ) )

How 'bout the patch below instead.


        Stefan


PS: Now that I think about it, maybe a better approach is to look for
    an alias that not only has a "." but additionally has the hostname
    as prefix.  This is what my second patch does.  Both patches are
    guaranteed 100% untested.


--- orig/src/sysdep.c
+++ mod/src/sysdep.c
@@ -2428,7 +2428,13 @@
 	    char *p;
 #endif
 
-	    if (!index (fqdn, '.'))
+	    if (!index (fqdn, '.')
+		/* If the IP address is 127.*.*.*, don't bother looking for
+		   a fqdn in the aliases since the name itself won't matter
+		   much and we may end up with localhost.localdomain
+		   or some other meaningless "fqdn".  */
+		&& !(hp->h_addrtype == AF_INET
+		     && hp->h_addr[0] == 127))
 	      {
 		/* We still don't have a fully qualified domain name.
 		   Try to find one in the list of alternate names */



--- orig/src/sysdep.c
+++ mod/src/sysdep.c
@@ -2428,8 +2428,16 @@
 		/* We still don't have a fully qualified domain name.
 		   Try to find one in the list of alternate names */
 		char **alias = hp->h_aliases;
-		while (*alias && !index (*alias, '.'))
-		  alias++;
+		while (*alias)
+		  {
+		    char *i = index (*alias, '.');
+		    /* hp->h_name is supposed to be the "official name" so we
+		       should not take any random fully qualified alias.  */
+		    if (i && strlen (fqdn) == i - *alias
+			&& !strncmp (fqdn, *alias, i - *alias))
+		      break;
+		    alias++;
+		  }
 		if (*alias)
 		  fqdn = *alias;
 	      }

Re: cvs emacs and (system-name)

[Stefan Monnier]

>>>> I don't really have a domain I would put my machine in, and I think
>>>> a reasonable number of people may fall into this category.  Emacs
>>>> picking "localhost.localdomain" over the actual name of my machine
>>>> still seems a bit odd.

Stefan> But it's kind of difficult for Emacs to guess which of the
Stefan> three names you put there is "my machine name".

> My recollection is that traditionally the first host name is
> considered the canonical one.  'man 5 hosts' backs me up on this :)

I believe the problem is that it is common to see entries in arbitrary order
in there.  Especially the fully-qualified entry is not necessarily first
(although it should).

> I don't have any other entry in /etc/hosts.  My machine has no fixed
> address and is not in a domain.  This, I believe, is extremely common
> among laptop users.

And in that case it's indeed rather unimportant which name Emacs chooses.

> Perhaps if only "localhost" and "localhost.localdomain" are found,
> Emacs could fall back to gethostname(2) or uname(2).  With glibc the
> latter appears to try to return the domain name as well.

What I was suggesting (tho implicitly) is to do that whenever the IP address
is 127.*.*.*.


        Stefan

Re: cvs emacs and (system-name)

[Chong Yidong]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> ! 		while (*alias && ( !index (*alias, '.') ||
>> ! 		     !strcmp (*alias, "localhost.localdomain" ) ) )
>
> How 'bout the patch below instead.
>
> PS: Now that I think about it, maybe a better approach is to look for
>     an alias that not only has a "." but additionally has the hostname
>     as prefix.  This is what my second patch does.  Both patches are
>     guaranteed 100% untested.

Is there any reason to think that we need to solve this for Emacs
22.1?

Re: cvs emacs and (system-name)

[Glenn Morris]

Chong Yidong wrote:

> Is there any reason to think that we need to solve this for Emacs 22.1?

No. Emacs 21 behaves the same way as 22 in this regard, so this is not
an urgent issue.

Re: cvs emacs and (system-name)

[Richard Stallman]

    At the moment, we prefer any name with a "." in. We could skip
    "localhost.localdomain":

That is a good idea.  Please do it.

Re: cvs emacs and (system-name)

[Glenn Morris]

Richard Stallman wrote:

>     At the moment, we prefer any name with a "." in. We could skip
>     "localhost.localdomain":
>
> That is a good idea.  Please do it.

Stefan had some other suggestions, I'll wait till you read those.
Emacs 21 behaved the same way, so it is not urgent we change this
and it should probably wait till after the release.

Re: cvs emacs and (system-name)

[Richard Stallman]

    PS: Now that I think about it, maybe a better approach is to look for
	an alias that not only has a "." but additionally has the hostname
	as prefix.  This is what my second patch does.  Both patches are
	guaranteed 100% untested.

This patch does seem likely to be really right, but it has the potential
to cause problems in peculiar cases.  For now, please install your limited
patch that rejects just localhost.localdomain.

After the release, please install the newer "more correct" patch.

Re: cvs emacs and (system-name)

[Chong Yidong]

Richard Stallman <rms@gnu.org> writes:

>     PS: Now that I think about it, maybe a better approach is to look for
> 	an alias that not only has a "." but additionally has the hostname
> 	as prefix.  This is what my second patch does.  Both patches are
> 	guaranteed 100% untested.
>
> This patch does seem likely to be really right, but it has the potential
> to cause problems in peculiar cases.  For now, please install your limited
> patch that rejects just localhost.localdomain.
>
> After the release, please install the newer "more correct" patch.

OK, checked in.