Bug 761877 (GLUSTER-145)

Lets track the progress here.

------------------
From: Shehjar Tikoo <shehjart>
Date: Wed, Jul 15, 2009 at 6:07 AM
To: dl-hacking <dl-hacking>


Hi Folks

Here is the summary of the points discussed in the meeting a few mins
back. The meeting dealt with the items to be added to the personal tasks
list of developers for the NFSv3 xlator effort.

Translator Responsibilities
===========================
AFR - Vikas
DHT - Avati
Posix - Shehjar
RA,WB - Raghavendra
Unify,io-cache - Gowda

Steps towards NFSv3 support
===========================
1. Create new branch.

2. Change fop prototypes: This step requires one person to go around
changing the prototypes in all xlators as required. The change is
cosmetic only because the change does not require that these new
arguments be interpreted by the fops.

Delegated to Shehjar
Deliverable: One patch per fop. This will ensure that every single
changed fop will still continue to result in a successful build.

3. Logic changes in xlators: Requires the above translator delegates
to start using those new arguments in the changed prototypes.

Deliverable: one patchset per xlator.

4. Merge to mainline.

By tomorrow(16.07, IST), the delegates will determine the time
required for implementing their tasks. Based on that, Vijay will discuss
how to schedule development.

-Shehjar

----------
From: Amar Tumballi <amar>
Date: Wed, Jul 15, 2009 at 9:01 AM
To: Anand Avati <avati>, Anand Babu Periasamy <ab>
Cc: dl-hacking <dl-hacking>


Avati/AB,

Do we need 'unify'  at all in 2.1?? I think maintaining all these changes and keeping it out of bugs for a module with no users, the effort is not worth. 

Only reason for unify to survive 2.0.x release is, DHT was unable to provide 'switch' support properly, but with the later readdir changes, implementing a switch behavior in DHT has become very simple. Hence I propose that we should be droping unify from the code base.

----------
From: Anand Babu Periasamy <ab>
Date: Wed, Jul 15, 2009 at 9:07 AM
To: Amar Tumballi <amar>
Cc: Anand Avati <avati>, dl-hacking <dl-hacking>


I think we should drop it. If you really have a reason to have it,
we should create classify it under deprecated/cluster/unify.
If it doesn't work at all in 2.1 and too much work to fix it, we should drop
it.

--
Anand Babu Periasamy
GPG Key ID: 0x62E15A31
Blog [http://unlocksmith.org]
GlusterFS [http://www.gluster.org]
GNU/Linux [http://www.gnu.org]

----------
From: Anand Avati <avati>
Date: Wed, Jul 15, 2009 at 9:46 AM
To: Shehjar Tikoo <shehjart>
Cc: dl-hacking <dl-hacking>


This is a good plan. What about the changes to the remaining translators?

Avati

----------
From: Vijay Bellur <vijay>
Date: Wed, Jul 15, 2009 at 10:05 AM
To: Anand Avati <avati>
Cc: Shehjar Tikoo <shehjart>, dl-hacking <dl-hacking>


Anand Avati wrote:
The translators that require extra functionality/logic to be added have been mentioned above. The remaining translators would have change only in prototypes and that would be done by  Shehjar. Do you foresee more changes in other translators?

-Vijay

----------
From: Anand Avati <avati>
Date: Wed, Jul 15, 2009 at 11:51 AM
To: Shehjar Tikoo <shehjart>
Cc: dl-hacking <dl-hacking>


Note that you would also have to change all the STACK_WIND and UNWIND invocations of those fops to pass extra dummy parameters, and make sure those dummy parameters are kept track of somehow while fixing the translators (by forcing a warning of some sort), and not get left behind in the end. To make things worse, our current STACK_UNWIND has no type check :)

Avati

----------
From: Amar Tumballi <amar>
Date: Wed, Jul 15, 2009 at 12:48 PM
To: Anand Avati <avati>
Cc: Shehjar Tikoo <shehjart>, dl-hacking <dl-hacking>


Avati, and Team,

I propose the idea of changing xlator.h first (for APIs), and then starting configure.ac from fresh set of translators which gets these changes properly.. least disruptive, and we have less dependency on the step 2 (that is one man job to change 1040+ STACK_WINDs, 1400+ STACK_UNWINDs, 930+ function definitions, and 1300+ fop_cbk definitions. I guess it will take minimum 14 days for one man if we start with cosmetic changes as first step. 

-Amar