Ticket 2260

Thomas, I'm a little concerned about putting this in 15.08 as it appears to change format.  What are your feelings on the matter?

Anyone using sh5util extract functions to generate csv files will not have any trouble.
But you are right, if any script/code use directly the h5 file usingan API can have some surprises.

It's ok for me to put this on 16.05.

Thanks Thomas, this has been added to 16.05 in commit 75317972f00.  I did change your %d to a %u though as nodeid is an uint_32.

Please reopen if anything else is needed on this.

Thomas, I have found this patch breaks sh5util (that is still trying to do things with the node_name instead of the nodeid).  I am thinking we should just add "Node Index" to the group node and revert this commit.

What was the purpose for the change in the first place?

I made the following commits that do what I proposed.

abcf2580f7c
91d63d9248b

I also did a minor rewrite of sh5util in commits

34f7ab28d95
ec25747c5f5

I would still like to know what the original reason behind the change was though.

Hi Danny,

The original reason was the loop done reading 'h5dump job_1.h5'
It is really easier to loop on node ids (as step ids) than on node names.

Note it is also not usefull to have 'ATTRIBUTE "Node Name"' with the value equal to the group name.

I haven't used h5dump before, only hdfview.  As noted sh5util was broken with the patch, I will also note the group on node id made looking at the tree with hdfview confusing as it only displayed the index instead on the node name for the group (as expected).  I'll go try h5dump and try to see what you are talking about, but as it stands today the patch would have to make sh5util work as well which it doesn't.  Currently I'm in favor of keeping it the way it is.  The view with hdfview wasn't as helpful when displaying the index instead of the name, I'm not sure many people would like that who use hdfview.  Perhaps my opinion will change when I look at it with h5dump, I'll let you know. 

On March 2, 2016 2:27:24 AM PST, bugs@schedmd.com wrote:
>https://bugs.schedmd.com/show_bug.cgi?id=2260
>
>--- Comment #6 from Thomas Cadeau <thomas.cadeau@bull.net> ---
>Hi Danny,
>
>The original reason was the loop done reading 'h5dump job_1.h5'
>It is really easier to loop on node ids (as step ids) than on node
>names.
>
>Note it is also not usefull to have 'ATTRIBUTE "Node Name"' with the
>value
>equal to the group name.
>
>-- 
>You are receiving this mail because:
>You are watching all bug changes.

After looking at it in h5dump I still see no reason to change the format as purposed.  I do see the nodename is redundant, but I don't think that is a big deal.  I suppose we could remove nodename attribute from the group, but I don't think it adds that much weight.  I like the group named after the node name instead of the node index.  I am not sure I see your concern yet.  Is this requested from a user?  Or are there other reasons to make the switch?

The request came from our validation team.
I agree, it doesn't add weight.

It should be intersting to have other users view and to know have they use it.
From our dev team's tests, we didn't see any problem to have the group named after the node.

It would be interesting.  I am not sure how it would help as to a user seeing 0 instead of the hostname I would expect to be less informative.  If you have a good use case in favor of the change please reopen this bug.