cvs commit: src/sys/conf files src/sys/netinet tcp_ofld.c tcp_ofld.h tcp_var.h toedev.h src/sys/sys socket.h

[Robert Watson]

On Thu, 13 Dec 2007, Scott Long wrote:

>> Let's not discourage that just yet.
>
> Yes, I would like to discourage disrespectful nit-picking of an important 
> piece of work.

I think you're reading too much into Bjoern's comments.

>> I'd like to see all significant changes to TCP discussed on public mailing 
>> lists well before they are committed -- at that point, someone saying 
>> "actually, I'd name the files a bit differently" is a lot easier to deal 
>> with than, say, immediately after they are committed. This needs to be 
>> communally owned and maintained code, or in two years time we'll find 
>> ourselves in the same position: architectural well-meant changes that are 
>> mostly right, but with no review of the details leading to the inevitable 
>> failures.
>
> A failure of what, exactly?  Will the names that Kip chose lead to failures 
> of TCP sessions?  Please enlighten me here.

This thread is a symptom of a specific problem: a failure to seek review for 
the work before committing.  I'm sure I'm not the only person who saw this 
commit and went, "So where was the public request for review for a major 
change to our TCP stack?"  Requests for more consistent naming, etc, are 
coming out now precisely because that review wasn't sort *before* committing. 
TOE represents a significant architectural modification, including a new KPI 
for device drivers to implement: details matter.  Some of these new filenames, 
function names, field names, etc, will be embedded in third-party source code 
for the forseeable future.  No one is saying that Kip's work isn't appreciated 
or valued -- rather, that at some point with a piece of code as sensitive and 
critical at TCP, it needs to go through careful review and refinement.  I sent 
Kip a large patch within an hour of his commit to clean up similar sorts of 
problems within the file,s making it comply more with the general TCP style 
but also to follow conventions for field-naming in data structures, etc, which 
he committed along with refinments of his own.

And, FWIW, this doesn't appear to be a bikeshed, because other than you 
arguing that this is turning into a bikeshed, no one seems to disagree with 
the proposed renaming so far.

Robert N M Watson
Computer Laboratory
University of Cambridge