Jan Harkes <jaharkes_at_cs.cmu.edu> writes:

> On Sat, May 19, 2007 at 04:51:07PM +0200, Enrico Weigelt wrote:
>>
>> So should we drop the openssl dependency completely ? 
>
> I think so, the openssl implementation may or may not be faster. The
> md5 and sha1 code is conditionally compiled in coda/lib-src/base. Every
> Coda binary links against the resulting libbase.a.

In general I would favor requiring a crypto library rather than having
included crypto code.  But openssl may be problematic - is the license
GPL-compatible?

> Of course we will lose the client-install/server-install targets, so it
> would be benificial to move that decision under a configure option but
> that way we can limit what is built when someone only wants to install
> only a client or a server.

But these targets don't really do the right thing.  My view (expressed
before) is that while it's good to have things set up easily for
building from source, the most important target is binary packaging
systems.   For those, client-install/server-install were never the right
thing, because there was no way to separate out what is common.

In an automake world, it's easy to conditionally define
@COMMON@
@CLIENT@
@SERVER@
to be '#' or '', and to be able to turn them on/off.  Certainly having a
way to install common and client is nice for source builders, but
packaging systems like pkgsrc want to be able to install just common, and
then on separate builds install just client and just server.

This could probably be done now, but the Makefiles seem messy and
fragile.