Subject: Re: Removing GNU tar and GNU cpio from src?
To: NetBSD Userlevel Technical Discussion List <>
From: James K. Lowden <>
List: tech-userlevel
Date: 01/25/2003 14:42:37
On Mon, 20 Jan 2003 15:16:17 -0500 (EST), "Greg A. Woods"
<> wrote:
> [ On Monday, January 20, 2003 at 09:54:29 (+0100), Xavier HUMBERT wrote:
> ]
> > Subject: Re: Removing GNU tar and GNU cpio from src?
> >
> > The whole point is that legacy tar could be used as a backup tool
> No, the point is that you were creating broken backups without knowing
> it.
> The whole point of backups is to be sure they record the system in a
> consistent state regardless of whether you actually verify them or not.

[I probably know better than to jump in here, but oh, well.]

	When I choose a word, it means just what I choose
	it to mean -- neither more nor less.
		-- Humpty Dumpty

You're choosing your definitions such that you can't be wrong.  I think
you should admit other definitions.  Xavier wrote:

>> But I don't really care for the very few
>> last lines of /var/log/news, while I certainly want to have the
>> remaining of my /var partition dumped. 

He doesn't want a perfect pristine backup from single-user mode; he wants
a best-effort backup of a running machine.  He's trading off data loss for
availability.  That's only "broken" if it doesn't meet his requirements. 
His users and his environment define those requirements; it's his neck at
risk.  No passage from the Church of Greg :-) will change that.  

I guess my definition of "broken" is usually the same as "has a bug" i.e.,
"behaves in exception to the documentation", but sometimes it can mean,
"lacks a feature required for a particular use".  In that latter sense,
pax could be said to be broken, because it's an incomplete replacement for
Gnu tar, one that, as things now stand, will require NetBSD sysadmins the
world over to adjust their scripts and/or install Gnu tar.  

>> At least, I'd like to have a return value for non-fatal errors, the
>> current version returns only 0 or 1 

He's only saying that he'd like nonfatal problems to yield a nonfatal
return code, because he thinks some picture of /var is better than no
picture of /var.  If anyone who thinks programs need more than two return
codes is Wrong, then I plead Guilty with an Explanation, your honor.