Mercurial repo for openjdk6 drops

[Kelly O'Hair]

Speaking from experience, any attempts at cleanly creating changesets
from TeamWare data alone is a waste of time. Just doesn't work well.
Granted we have pretty complete OpenJDK6 changeset data (data beyond
the TeamWare data) and could potentially do this, it's quite a bit of
grunt work to accomplish. Before we go off on this kind of adventure
we need to make sure the time spent is worth it.
A different approach would be to create changesets per build/promotion,
but those kind of large changesets can have limited usefulness.

I'll try and talk to Joe about this tomorrow.

Also, don't forget that the OpenJDK7 sources are a Mercurial forest,
not just one repository. I assume that we want to mimic the same
forest layout, keeping the langtools, corba, jaxp, jaxws, and hotspot
sources in their own repositories, separate from the core 'jdk' repository.
That allows us to do some plug and play with these repositories
between OpenJDK6 and OpenJDK7.

-kto

Mark Wielaard wrote:
> Hi Joe,
> 
> On Tue, 2008-09-16 at 09:37 -0700, Joseph D. Darcy wrote:
>> While I'm sympathetic to the request and appreciate the offer of help, 
>> it is unlikely I'll be able to publish the initial Mercurial 
>> repositories with any history.  The most history that could be published 
>> dates back to when OpenJDK 6 branched off from the already open sourced 
>> JDK 7.  The total amount of bugs fixed from then until now is on the 
>> order of 400-500, with somewhat fewer logical changesets.  This number 
>> of changesets is quite small compared to the amount of pre-Mercurial 
>> history that is not being published externally at all.
> 
> That is actually about half of what jdk7 or icedtea (disregarding
> openjdk import changes) have been seeing. So I do think it is a
> significant amount of changes. It would certain help with getting a
> better view of how changes relate to each other.
> 
>> However, the 
>> number is large enough that is would be troublesome for me to reliably 
>> synthesize changesets.   I don't have Martin's process for scripting and 
>> SCM hackery and one of the limitation of the teamware SCM is that the 
>> changes are only really tracked on a per-file basis.  Published 
>> truncated teamware histories outside of Sun's firewall for people to 
>> help cons up changesets might be technically possible, but much of the 
>> prep work and reviews would have to be done internally and the 
>> costs/benefits on that portion don't look attractive to me.
> 
> Ah, that is a bummer. I had assumed that the teamware history only
> included stuff starting with the public release of openjdk7. Isn't there
> any way to "replay" the patches starting at that point and post those?
> rcs and cvs also maintain history on a per-file basis, but with some
> scripting tricks it is still possible to reconstruct cross-file
> changesets from them mostly. If that is impossible, could you start
> posting patches to the jdk6-dev when you apply them to your repository?
> So the chunks are smaller and can be commented on more timely than when
> you do full bxx code drops?
> 
> Thanks,
> 
> Mark
>