Re: [tlug] SVN: Your Environment, BerkleyDB/Web or FSFS?

[Edward Middleton <edward@example.com>]

the following two emails will probably give more insight into the issues
that arose developing subversion.

http://web.mit.edu/ghudson/thoughts/diagnosing
http://web.mit.edu/ghudson/thoughts/undiagnosing


Ian Wells wrote:

> So my argument would be: why not find a way of serving DAV standalone

Because subversion adds a lot to apache.  The standard web_dav modules
has no versioning backend.  It is not much better then ftp.  Just by
being an apache module it exposes itself to a much larger audience (more
testing).

> rather than writing a whole new protocol and adapters for both ends? 
> With cheats, if necessary - it can still be a non-DAV protocol, but
> the aim is to share the code.

I think it was the other way round.  I understand it was much easier
implementing the svn protocol then the web_dav/DeltaV protocol.

> Not saying it's feasible.  It just looks like a better solution to me
> long term,

I agree one protocol is going to be better then two.  I believe the
problem was that the required features of web_dav hadn't standardized
when subversion needed them so rather then wait for the standards to
catch up (or hack some propriatry extension ms style) they made a
propriatry protocol with the web_dav interface as a secondary to be done
we the standard situation settled.

> because you have one protocol and everybody's finding the bugs in it
> for you, rather than two protocols and twice the bugs, and an audience
> split between them so that each is being tested only half as much. (*)

 I believe the versioning code was actually the much harder/larger part
of the development.  By having two protocols that shared the same
versioning components they could satisfy diehard cvs users who wouldn't
run no f'ing apache server (there were actually a lot of people who
would not consider using subversion because they thought it required
apache http://subversion.tigris.org/faq.html#need-apache) running on
there SCM box and they could leverage the fact that apache's web_dav did
90% of a web dav protocol for them.

> (* For the pedants: agreed, arguably "use" isn't "testing".  In which
> case divide the testers between the code rather than the users. 
>
> What, you mean your OSS project has no testers?)

I gather the developers are big on unit testing but can't find any
references.

Edward