[quagga-dev 8817] Re: [PATCH 1/2] lib: Optimize Fletcher and Internet checksums

Paul Jakma paul at jakma.org
Thu Sep 8 13:52:37 BST 2011


On Thu, 8 Sep 2011, Joakim Tjernlund wrote:

> Just found out that the odd byte handling in my patch is wrong for BE 
> archs(even the RFC got this wrong apparently)

So I was justified in not accepting this patch because it had no unit 
test? :)

Seriously though, patches need to come with unit-tests as much as 
possible. Certainly, there's little reason code can not be organised so 
that functional logic ('functional' in the "not dependent on state other 
than the arguments" sense) is separated as much as possible from the 
stateful logic. Then there should be no reason why the functional parts 
at least can not be supplied with unit-tests.

Also, we need to get away from subjective justifications for patches, and 
instead quantifiably measure their impact, if we want to get to an 
objective basis for committing patches. So if there's a performance 
justification for a patch, there ought to a test-case (preferably an 
isolated one) to demonstrate the change in performance.

In summary, we need to get to a place where the norm for code 
contributions is that they:

* Separate out functional logic, for ease of testing

* Come with unit-tests

* Come with an objective justification for their integration (e.g. fixes X
   bugs, comes with a Y% perf improvement on this test-case).

Also, on a tangent, I've long gotten got quite sick of writing commit 
messages for other people.

regards,
-- 
Paul Jakma	paul at jakma.org	Key ID: 64A2FF6A
Fortune:
It was Penguin lust... at its ugliest.



More information about the Quagga-dev mailing list