Bug 58631
| Summary: | <synoppossibly poor default in noattach's Makefile | ||
|---|---|---|---|
| Product: | Ports & Packages | Reporter: | Luigi Rizzo <rizzo> |
| Component: | Individual Port(s) | Assignee: | Dirk Meyer <dinoex> |
| Status: | Closed FIXED | ||
| Severity: | Affects Only Me | CC: | dinoex |
| Priority: | Normal | ||
| Version: | Latest | ||
| Hardware: | Any | ||
| OS: | Any | ||
Responsible Changed From-To: freebsd-ports-bugs->dinoex Over to maintainer State Changed From-To: open->feedback Please use: SENDMAIL_MILTER_IN_BASE=yes its ecxacly for your needs. You can put this in /etc/make.conf So far only 4.8 RELEASE and 5.1 RELEASE have an recant version of libmilter. see http://www.freebsd.org/cgi/cvsweb.cgi/src/contrib/sendmail/libmilter/ This ports links in the latest version of libmilter.a I see no reason to change this. On Tue, Oct 28, 2003 at 10:13:01PM +0100, Dirk Meyer wrote: A ... > Please use: > SENDMAIL_MILTER_IN_BASE=yes I know this -- I am just saying that the default seems to require an unnecessary build of mail/sendmail when the Makefile could easily detect the existence of the required library in the standard places, and force the dependency only if not available. E.g. something like - .if !defined(SENDMAIL_MILTER_IN_BASE) + .if !exists(/usr/lib/libmilter.a) BUILD_DEPENDS= ${LOCALBASE}/lib/libmilter.a:${PORTSDIR}/mail/sendmail .endif (actually, it should check for the dynamic version, libmilter.so.2) and then decide which library to use (if you have it both in /usr/lib and ${LOCALBASE}) with the order of directories in LD_LIBRARY_PATH cheers luigi > its ecxacly for your needs. > You can put this in /etc/make.conf > > So far only 4.8 RELEASE and 5.1 RELEASE > have an recant version of libmilter. > > see > http://www.freebsd.org/cgi/cvsweb.cgi/src/contrib/sendmail/libmilter/ > > This ports links in the latest version of libmilter.a > I see no reason to change this. > > > http://www.freebsd.org/cgi/query-pr.cgi?pr=58631 Luigi Rizzo schrieb:, > I know this -- I am just saying that the default seems to require > an unnecessary build of mail/sendmail when the Makefile could easily > detect the existence of the required library in the standard places, > and force the dependency only if not available. E.g. something like > > > - .if !defined(SENDMAIL_MILTER_IN_BASE) > + .if !exists(/usr/lib/libmilter.a) > BUILD_DEPENDS= ${LOCALBASE}/lib/libmilter.a:${PORTSDIR}/mail/sendmail > .endif This will break all installations for Users not running CURRENT or STABLE. Even if /usr/lib/libmilter.a does exist it is not the prefered library. an old "/usr/lib/libmilter.a" might work due the interface is stable. But it does not contain the buxfixes up to the latest version. > (actually, it should check for the dynamic version, libmilter.so.2) > and then decide which library to use (if you have it both in /usr/lib > and ${LOCALBASE}) with the order of directories in LD_LIBRARY_PATH SENDMAIL_MILTER_IN_BASE=yes will use the dynamic version it exists. kind regards Dirk - Dirk Meyer, Im Grund 4, 34317 Habichtswald, Germany - [dirk.meyer@dinoex.sub.org],[dirk.meyer@guug.de],[dinoex@FreeBSD.org] your logic is that libmilter from the ports is preferable to the one that exists in the installed system. However this is only true if the port tree is updated, which is equally On Wed, Oct 29, 2003 at 04:44:45AM +0100, Dirk Meyer wrote: > Luigi Rizzo schrieb:, > > > I know this -- I am just saying that the default seems to require > > an unnecessary build of mail/sendmail when the Makefile could easily > > detect the existence of the required library in the standard places, > > and force the dependency only if not available. E.g. something like > > > > > > - .if !defined(SENDMAIL_MILTER_IN_BASE) > > + .if !exists(/usr/lib/libmilter.a) > > BUILD_DEPENDS= ${LOCALBASE}/lib/libmilter.a:${PORTSDIR}/mail/sendmail > > .endif > > This will break all installations for Users not running CURRENT or STABLE. > Even if /usr/lib/libmilter.a does exist it is not the prefered library. > an old "/usr/lib/libmilter.a" might work due the interface is stable. > But it does not contain the buxfixes up to the latest version. If what you are trying to avoid is the use of an existing but stale libmilter.* then you should _check_ that the existing library is not good, not just _assume_ that the one in /usr/lib is bad and the one in ${LOCALBASE}/lib is good. Both assumptions (well, assumptions in general) are flawed. Your very same reasoning about a not-updated library applies to the port as well -- in the above dependency you don't ask for sendmail-X.Y.Z but just for any sendmail, so it might well be the same one that got installed when the user fetched the ports tree. In fact, if you want to deal with users that do not track the base source tree, why do you assume that they he track the ports tree instead (risking all sorts of incompatibilities because most ports do not properly check versions of the installed software) ? I hope i made my point clear, which is, i believe the makefile does something unnecessary for "users not running current or stable". I know you have put in tweaks for those users, i find it weird that you consider them the "non default" case. cheers luigi > > (actually, it should check for the dynamic version, libmilter.so.2) > > and then decide which library to use (if you have it both in /usr/lib > > and ${LOCALBASE}) with the order of directories in LD_LIBRARY_PATH > > SENDMAIL_MILTER_IN_BASE=yes will use the dynamic version it exists. > > kind regards Dirk > > - Dirk Meyer, Im Grund 4, 34317 Habichtswald, Germany > - [dirk.meyer@dinoex.sub.org],[dirk.meyer@guug.de],[dinoex@FreeBSD.org] > your logic is that libmilter from the ports is preferable to the > one that exists in the installed system. However this is only > true if the port tree is updated, which is equally > On Wed, Oct 29, 2003 at 04:44:45AM +0100, Dirk Meyer wrote: [...] Users do use portupgrade/portseasy and others tools. The need to cvsup to ports tree as whole, to get the latest version of noattach. > If what you are trying to avoid is the use of an existing but stale > libmilter.* then you should _check_ that the existing library > is not good, not just _assume_ that the one in /usr/lib is bad and > the one in ${LOCALBASE}/lib is good. It can be ensured that the version in the ports tree is greater or eaual the version in the base system. > In fact, if you want to deal with users that do not track the > base source tree, why do you assume that they he track the ports tree > instead 1) track the ports tree is commen use. 2) base source tree is often in sync with a RELEASE. e.G. resulting to "5.1-RELEASE-p10" > (risking all sorts of incompatibilities because most ports > do not properly check versions of the installed software) ? Please submit patches if you see room for improvent. > I hope i made my point clear, which is, i believe the makefile does > something unnecessary for "users not running current or stable". > I know you have put in tweaks for those users, i find it weird > that you consider them the "non default" case. sorry I can't trace your logic. - [dirk.meyer@dinoex.sub.org],[dirk.meyer@guug.de],[dinoex@FreeBSD.org] State Changed From-To: feedback->closed feedback timeout |
ports/mail/noattach/Makefile has the following: .if !defined(SENDMAIL_MILTER_IN_BASE) BUILD_DEPENDS= ${LOCALBASE}/lib/libmilter.a:${PORTSDIR}/mail/sendmail .endif but at least on RELENG_4 (and possibly also CURRENT) libmilter.a is already present in /usr/lib as part of the default installation. I believe the default should be to assume libmilter.a is present, and only if explicitly requested the dependence should be on ${LOCALBASE}/lib/libmilter.a:${PORTSDIR}/mail/sendmail