[quagga-dev 7714] Re: [PATCH] zebra: consider all nexthops when looking fora gateway match

Joakim Tjernlund joakim.tjernlund at transmode.se
Mon Feb 1 12:09:32 GMT 2010


>
> On Mon, 1 Feb 2010, Alexis Rosen wrote:
>
> > I'm completely in sympathy with the desire to clean up the code,
> > but refusing this patch won't make that happen any faster.
>
> > Reviewing it and accepting it would probably take less time than
> > responding in the negative multiple times a year as the issue comes
> > up, which is in fact what's been going on.
>
> Sigh.
>
> There have not been negative responses to that patch. To the contrary
> in fact.
>
> The problems are 2 fold, one relating to the RIB and one relating to
> "How are maintainers to know your patch is 'good'":
>
> 1 The RIB:
>
> It's a bit of a mess at present. People poke at it to fix small bugs
> every now and then, but it's clearly much harder to understand than
> it ought to be. Joakim's own patch makes that comment; patches to the
> RIB (including the one that initiated this thread) often go through
> more than 1 revision even when the fix is conceptually simple.
>
> Further, precisely because the code is a mess and hard to understand,
> it becomes hard for maintainers to apply patches. Which brings us
> to...
>
>
> 2. How can we maintainers judge whether a patch is 'good' or not?
>
> We tend to go by:
>
> a) Our ability to understand the fix
> b) Hard empirical evidence (i.e. the output of test cases; even
>     better: unit tests we can easily run ourselves)
> c) Anecdotal evidence from feedback on the list
>
> Now, the last one is an interesting one. Clearly, feedback is from a
> self-selected set - ones who have had the problem. From this, and
> somewhat confirmed by experience, it follows that feedback can not be
> relied upon to spot regressions.

Which is why a devel tree is needed. You will never get past the self-selected
set otherwise. The self-selected set is pretty much the end of the road for
non maintainers. After that one will just have to wait until some maintainer
"understands" the patch 100%, has test it himself properly. That doesn't scale
and this is what we have been seeing the last year(s?).

        Jocke




More information about the Quagga-dev mailing list