Re: CVS commit: src/sys/arch/xen/xen

[Jean-Yves Migeon <jeanyves.migeon%free.fr@localhost>]

On Wed, 6 Apr 2011 16:29:22 +0000, Taylor R Campbell 
<campbell+netbsd%mumble.net@localhost> wrote:
> Date: Mon, 04 Apr 2011 23:46:19 +0200
>    From: Jean-Yves Migeon <jeanyves.migeon%free.fr@localhost>
>
>    The newer scripts for Xen read the interface value from the 
> vifname
>    entry in Xenstore, so changing the name is not that critical. But 
> this
>    should be stabilized sooner than later. Personally, I find '_' 
> rather
>    ugly in an interface name... but I am open to suggestions, even 
> fixing
>    rc.d/network if it needs to.
>
> One could declare that ifnames must match [a-z][-a-z0-9]*[0-9], and
> then in /etc/rc.d/network pass them through `tr - _' before 
> evaluating
> `args=\$ifconfig_$int'.  This guarantees that setting ifconfig_ifN in
> /etc/rc.conf always works, and depends only on agreeing to the ifname
> policy and ensuring that all the existing ifnames actually follow it.
>
> Of course, tr might not be available yet in /etc/rc.d/network, but
> someone wizardlier (or more patient) with sh than I can probably work
> around that...

Yep, but there are other places where this will get tricky. For 
example, rc.conf(5) is parsed as a classic shell script [1]. Given that 
someone has two possibilities to configure interfaces:

/etc/ifconfig.xxx
or
ifconfig_xxx # in rc.conf(5)

If we follow the same "xxx" naming convention for both, we would have 
to 'tr' certain lines of rc.conf for interfaces that contains a 
[^a-zA-Z0-9_] char. This complexifies rc.conf parsing for no real 
benefit. Putting a ifconfig_xxx-yyy="auto" in rc.conf(5) will likely 
result in a parsing error.

I am about to revert my patch anyway, and use '_' instead. I will also 
update ifconfig.if(5) to clearly indicate that only chars accepted in 
shell variable names should be used.

[1] http://nxr.netbsd.org/xref/src/etc/rc#22

--
Jean-Yves Migeon
jeanyves.migeon%free.fr@localhost