Re: contrib/jemalloc

On 04/20/2012 11:18 AM, Pedro Giffuni wrote:
> FWIW,
> 
> While the vendor branch is usually the cleanest way to merge
> updates, it is not always the best. I personally gave up on
> updating two packages from the vendor tree because it's just
> too much trouble.

With due respect, if doing it the right way is too difficult, the answer
is to ask for help rather than giving up. There are plenty of us who are
experienced with doing this, and would be glad to assist.

In the CVS era I agree that vendor imports could be very difficult. But
with svn it's much, much easier.

> In this case it's likely that the committed
> jemalloc is very FreeBSD specific and doesn't really match the
> more generic version.

That's true of most of our 3rd party products. We usually handle this
with something like FREEBSD-Xlist, etc. Once again I put forward the
FREEBSD-Upgrade file for BIND as an example. (And lest people think that
I'm tooting only my own horn here, a lot of that work, especially the
initial work for BIND 9, was done by des.) It's hard for me to see how
Jason's multiple patching/diffing solution could be an improvement.

Also, one advantage to doing a clean import into a vendor branch before
merging into contrib is that with svn maintaining local differences in
contrib/ is painless. In that way, whatever differences there are can be
easily seen by people not as familiar with the code as Jason is.

> INHO, being that the committer is also the author it is likely
> his prerogative how to update it.

I'm more interested in long term maintainability, as well as the
historical value that having the clean vendor imports gives us.

Doug