Date: Wed, 20 Apr 2005 14:27:07 -0400 From: Yarema <yds@CoolRat.org> To: Oliver Lehmann <lehmann@ans-netz.de>, freebsd-ports@freebsd.org Cc: Jose M Rodriguez <josemi@freebsd.jazztel.es> Subject: Re: splitting courier-authlib into master+slave ports Message-ID: <9F0C7E4FBA18BFD0B3946DFE@tuber.coolrat.org> In-Reply-To: <20050420121254.34c59e53.lehmann@ans-netz.de> References: <20050414111426.775f6afd.lehmann@ans-netz.de> <200504191258.53327.freebsd@redesjm.local> <20050419173039.4725638a.lehmann@ans-netz.de> <200504191827.05896.ports@dino.sk> <20050420121254.34c59e53.lehmann@ans-netz.de>
next in thread | previous in thread | raw e-mail | index | archive | help
--On Wednesday, April 20, 2005 12:12:54 +0200 Oliver Lehmann <lehmann@ans-netz.de> wrote: > Milan Obuch wrote: > >> On Tuesday 19 April 2005 17:30, Oliver Lehmann wrote: >> > Jose M Rodriguez wrote: >> > > >> > > Hope this may be on time. Can you consider work this? >> > > >> > > - maintain couier-auhlib just as a metaport selector. >> > > - add a courier-authlib-base port >> > > - maintain the courier-authlib-method ports, but working against >> > > courier-authlib-base. >> > >> > That sounds cool. then OPTIONs could be put into the metaport. I'll >> > work on this. >> >> I like this idea too. Actually I was trying to formulate tha same. You >> were quicker :) >> Milan > > Can you try the tar.gz once more? I uploaded an "adjusted" version ;) FWIW I'd like to weigh in with my opinion. I think this move to a meta port just so we can have OPTIONS selectable dependencies does little to improve usability. As I've argued before in an email to Oliver there's little need to have more than one courier-authlib-method port installed unless one is transitioning from one auth-method to another or just experimenting. So why is it better to: % cd courier-authlib && make config && make install rather than just: % cd courier-authlib-METHOD && make install (Yes, I know the 'make config' is not necessary ;) my point still stands.) We're just adding unnecessary complexity just because we can. A pkg-message in the base port (without OPTIONS) is sufficient to indicate to the user that other method ports are available IMHO. The "adjusted" versions of courier-authlib as I presented it to Oliver can be found at <http://yds.CoolRat.org/freebsd/>; The most recent one one incorporates some of the adjustments Oliver made to his version. One difference between the courier-authlib-20050408.00.tgz version and courier-authlib-20050420.00.tgz is that I make --with-authpam part of the base port's CONFIGURE_ARGS. This prevents libauthpwd.so.0 from being built and instead builds lib/courier-authlib/libauthpam.so.0. authpwd is discouraged as per <http://www.courier-mta.org/authlib/README_authlib.html>: NOTE: It might be tempting to throw in a towel and use authshadow or authpwd if you cannot figure out how to install PAM support, however that is not advisable. It is highly recommended to use authpam wherever the PAM library is available. The authpwd module is also documented in the same README to use "the C library's getpw() functions" which in turn are documented to be made "made obsolete by getpwuid(3)" in the FreeBSD getpw(3) man page. So given the above two citations from both courier-authlib docs and FreeBSD's docs why not just do away with authpam being optional and make it the default part of the base package? Some notes on the how what I did differs from Oliver's version and why: USE_RCORDER= authdaemond.sh instead of RC_SCRIPTS_SUB= PREFIX=${PREFIX} RC_SUBR=${RC_SUBR} MAILUSER=${MAILOWN} MAILGROUP=${MAILGRP} ${SED} ${RC_SCRIPTS_SUB:S/$/!g/:S/^/ -e s!%%/:S/=/%%!/} \ ${FILESDIR}/courier-authdaemond.sh > ${WRKDIR}/courier-authdaemond.sh I rewrote and renamed ${FILESDIR}/courier-authdaemond.sh as ${FILESDIR}/authdaemond.sh.in because the SED idiom Oliver used is already in <bsd.port.mk> so there's no need to write it explicitly if one just assigns a filename to USE_RCORDER= without the .in suffix. Search for every occurrence of SUB_FILES in <bsd.port.mk> to see what I mean. Also the way I rewrote ${FILESDIR}/authdaemond.sh.in loads it at about the same spot on the RCORDER as cyrus-sasl. And using cyrus-sasl as a model why call it courier-authdaemond.sh when the executable is authdaemond? saslauthd.sh is not named cyrus-saslauthd.sh... In short my version of the startup script is smaller without taking away any checks or functionality and without duplicating code already in /etc/rc.subr Getting back to the Makefile.. Olivers current version has a RUN_DEPENDS= ${LOCALBASE}/share/sysconftool/sysconftool:${PORTSDIR}/devel/sysconftool regardless weather it's a base or method port. Since dependencies are inherited there's no need for this. And the major difference is mainly organizational.. I elected to construct PLISTs for all the method ports dynamically within the main Makefile rather than using a file for each. This allows for a rudimentary form of code reuse to generate the TMPPLIST and cuts down on the number of files that make up a port. The latter is a Good Thing considering how many inodes the ports tree uses already. :) And that post-install: @${GREP} '^@exec ' ${TMPPLIST} \ | ${SED} -e 's:^@exec ::' -e 's:%D:${PREFIX}:g' \ > ${WRKDIR}/.PLIST.exec \ && ${SH} ${WRKDIR}/.PLIST.exec that I like also helps to eliminate a lot of code redundancy. Otherwise one needs to be really careful to add the code producing the same results to both the Makefile and the PLIST file so that the pkg_add and 'make install' produce the same result. I find it easier to just use the above grep(1)/sed(1) idiom and put all the code in just one place, the PLIST. Not the prettiest solution but less cumbersome than duplicate code. Oliver, now that you've hacked at this for a while please take a look at the way I did things again. The OPTIONS/metaport stuff is just not woth the effort, me thinks. And it does nothing for the pkg_add user who never sees the menus, but might be even more confused by having even more ports to chose from. With authpam in the base port just installing courier-authlib should suffice for the majority of users. For the rest all they have to do is install the method port they need and the base gets sucked in as a dependency. Perhaps instead of courier-authlib being the meta port create a courier-authlib-meta or -menu and leave courier-authlib as the base. Helps preserve the consistency of suffixes used only for method ports in the Makefile code. And doesn't force anyone to use the OPTIONS/meta port. -- Yarema http://yds.CoolRat.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9F0C7E4FBA18BFD0B3946DFE>