[quagga-dev 9678] Meta (was: Re: [PATCH RFC] OSPF vertices memory exhaustion)
equinox at opensourcerouting.org
Mon Aug 20 14:13:23 BST 2012
Ooooh, meta maintaining!
On Wed, Jul 25, 2012 at 09:07:44PM +0100, Paul Jakma wrote:
> It seems you didn't convince me this was wrong before. I've just gone
> through the original bug and refreshed my understanding from scratch as to
> why that was done as it is, and I still don't think it's wrong. Indeed,
> I'm not convinced you understand why that code is the way it is (based
> just on the commit comment, and my own refresh of the code, the blog
> entry, and bug #330 - I havn't gone over earlier discussions). ;)
> Now, that's not meant as an insult, Just that my approach to maintenance
> was that the maintainer should take a sceptical stance on contributions
> and require that they come with convincing explanations as to what the
> problem is, and why the submitted patch fixes it. Otherwise, I felt then,
> we risked lots of pointless code churn from speculative or
> not-fully-understood patches.
> That said, maintenance approach today may be different.
The approach I've been taking so far has been "merge by default".
However, this has caused a few bits here and there, in particular:
- I've picked up Roman's CLI address parsing patch, only to notice on
processing Juliusz' patch that Roman's patch partially breaks command
- Jorge's BGP cleanups first got two immediate fixups and now seem to
be responsible for the assert I'm seeing. The gain vs. regressions
was a bit low on that series TBFH. (Sorry Jorge...)
I had hoped that picking up patches with regressions would be seldom
exceptions, this doesn't seem to hold. Bit of a shoe in mouth moment
for me to be honest.
That said, I haven't given up on the approach yet. I would like to take
this as an opportunity to mention that if anyone has been using master
as a "rolling release", now would be a good time to stop that.
Let's put this in caps: GIT MASTER IS NOT AN USABLE QUAGGA VERSION.
The whole idea and enabling factor for the permissive merging approach
is that we have the OSR testing. I fully expect that we'll find another
regression or two there. If it turns out that permissive merging is
lowering the quality *too* much, we'll have to rethink this.
(This was also a big factor for me in picking up Joakim's OSPF changes.
For what I know, it might actually make the router melt -- I reviewed
and understood the patches, but I didn't prove them correct.)
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 230 bytes
Desc: Digital signature
More information about the Quagga-dev