DragonFly kernel List (threaded) for 2003-09
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]
Re: Anybody working on removing sendmail from base?
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On Friday 26 September 2003 12:39 pm, Matthew Dillon wrote:
> The way I envision mail replacement is to create an 'API' through
> an RCNG script which 'mail system ports' would have to conform to.
>
<snip>
> This is just an example of how it could work. In anycase, this would
> provide a migration path for sendmail to leave the base system because
> the RCNG support could be worked on right now, without removing
> sendmail, then the sendmail port could be modified to use the new API and
> tested (by using mta_sendmailXXX_* RC variables and turning off the base
> sendmail with the original RC variables), and then as a final step sendmail
> could be removed from the base system).
This same framework could be used for other stuff (bind) and more significant
bits as well, such as gcc and perl. The fun thing in those cases would be
ensuring that, e.g., /usr/bin/cc is symlinked to /etc/rc.d/cc, to determine
the correct CC to run (you don't want to compile the kernel with the wrong
version of gcc, it might not work, which is probably why the other BSDs
included it in the base system, while making ports available for newer
versions. On the other hand, sometimes it might work, and it would be nice
to play with it and see)
This scenario, however does raise another interesting question, applicable to
the portsng framework: currently as part of buildworld, gcc, sendmail, etc
are rebuilt from scratch (which makes sense, since the version of gcc used
from one version to the next may change) Does this mean that portsng will be
rebuilt during buildworld? Or will only certain parts, the "important bits"
such as gcc be rebuilt? Or will we be able to specify which parts to build
or not build? I ask this becuase of two potential issues. First, under the
current (FBSD) ports system, a 'make clean' removes all files from the ports'
directory, where buildworld would want the source present to be able to
rebuild--especially in cases where the port version has changed, but we
haven't updated, so the source for our version may no longer be available.
(for example, I usually keep my ports TREE up-to-date, so fetching the
distfile may not have the version I am expecting--unless we do what FBSD has
done with gcc and make 6 different versions available in ports). The other
problem with simply rebuilding all ports is that then you are looking at, in
some cases at least, rebuilding a lot of very big ports (Xfree86), which are
quite time consuming, thus dramatically increasing buildworld time. I guess
ideally there would be a way to specify which ports we want to rebuild during
buildworld. On the flip side of this is the fact that it probably isn't
always strictly necessary to rebuild gcc (not to mention sendmail, bind, etc)
as part of a buildworld, unless we want to update something, or the new
kernel/world wants a newer gcc--which could be specified in the makefile.
Well, enough rambling for a while....
mike
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (FreeBSD)
iD8DBQE/dbCXY30jZzkECLcRAiKjAJ9e9tmi06jp64U4EvMllubl15na6gCggN/Y
NgtlJMaUqh7WKId7vDXCYI8=
=tnko
-----END PGP SIGNATURE-----
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]