Re: [Radmind-devel] radmind support for extended file attributes

[Karl K. <kue...@gm...>]

On Aug 20, 2008, at 9:30 AM, Mark Montague wrote:

>> We also might just be running into the limitations of the current  
>> transcript format.
>
> Acknowledged and agreed: this is a problem.
>
> Besides xattrs and ACLs, what other things might be desirable to add  
> to transcripts in the future?  Were the transcript format to be  
> modified (as would need to happen for xattr support), it would be  
> good to take care of other needs at the same time.

	A while ago for these same reasons I suggested evaluating xar as a  
format. The xml-based catalog file would allow things like platform- 
specific ACL and other things (like other xattrs) in an straight- 
forward manner, and the whole system would allow future expansion  
without worrying about issues like this.

	And ACL's are going to have to be handled in platform-specific  
manner, as they appear similar but have lots of gotchas (especially if  
eventually this is going to bring Windows into the fold).

	The idea was shot down hard, mostly for misunderstandings that xar  
would have the same problems that tar would have. Since Radmind reads  
all transcript files linearly the problems sited are really non-issues  
(with the possible exception of case sensitive/insensitive sorting).  
In fact the random-access times on xar are good (since you only have  
to parse one section of the file to get the desired offset).

>> Secondly, adding support for xattrs will mean a performance hit. I   
>> can't say just how much yet, but for a desktop operating system  
>> like Mac OS X 10.5, which has hundreds of thousands of files in a  
>> default installation and suffers from weak I/O performance to begin  
>> with, the performance penalty is likely to be noticeable.
>
> Yes, but I would envision xattr support in radmind being turned on  
> only for clients where the system administrator wanted it.  Clients  
> without xattr support turned on could simply ignore any xattr or ACL  
> information that may be present in transcripts.

	And isn't one of the central ideas of Radmind to support security?  
ACLs are the way that is going on a lot of platforms. It is a hit on  
performance but the current cost in terms of security lost on the  
MacOS X side is already large (but manageable in default form).


--
		Karl Kuehn
			la...@so...