Hi Daniel, On Mon, Mar 16, 2009 at 6:01 PM, Daniel Veillard <veillard@redhat.com> wrote:
On Sun, Mar 15, 2009 at 02:47:14AM +0900, Ryota Ozaki wrote:
Hi,
This patch allows to specify owner/group of permissions in storage XML as alphabetical names.
One concern of this patch is since alphabetical names are converted to uid/gid at XML parsing (e.g., virsh pool-define some-pool.xml), the generated XML file for libvirt internal does not contain the original alphabetical names and dumpxml does not as well. Is it better the conversion is done when activated not defined?
Yes, I'm wondering about this. Usually it's better if the XML can round-trip without change (i.e. the exported version is the same as the imported one), modulo maybe some reordering of elements. But loosing information or distortions in the cycle tend to generate problems. So I'm wondering what is in your opinion the added value of having livirt do that conversion instead of the layer on top of libvirt.
The reason why I want this function is just ease of use because we usually specify owner/group by names not numbers, but I have no objection to your concern of loosing information. So I will revise the patch to fix the problem. And the reason I did odd implementation is that I worried about the impact by modifying the internal data structure (apologies I didn't tell this). How about this point?
There is the added problem that the given name might not be portable from one node to another, that's true too for uid/gid values but for example in the case of networked filesystems, the hardcoded numbers are actually more portable than the high level name equivalents.
I thought same thing, but I think it's probably not matter. In many cases, user accounts would be centrally managed, and if not so, hardcoded numbers also don't make sense because uid/gid values make sense only if the corresponding accounts exist. So IMO the portability thing should be considered in operations.
So I'm not sure really if the advantage really outweight clearly the potential problems. I don't know, it's probably a matter of use cases.
Anyway, I'm sure I will remain the capability to specify with numeric numbers, so there should be no regression ;-)
Otherwise the patch looks fine to me, test case may be harder due to the non-portability of the names.
Thanks for reviewing! and I agree with you. If writing test cases, we need a trick, for example, creating a temporal account that has very higher unused uid/gid only for the tests. Thanks again, ozaki-r
Daniel
-- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel@veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/