Portmaster and Portmanager problem with jdk15

[Robert Noland]

On Thu, 2007-08-09 at 22:07 +0100, RW wrote:
> On Thu, 09 Aug 2007 12:41:06 -0400
> Robert Noland <rnoland at 2hip.net> wrote:
> 
> > On Thu, 2007-08-09 at 09:47 -0600, Greg Lewis wrote:  
> > > On Tue, Aug 07, 2007 at 04:59:09PM -0400, Robert Noland wrote:  
> > > > On Tue, 2007-08-07 at 13:28 -0700, Doug Barton wrote:  
> > > > > RW wrote:  
> > > > > > Both Portmaster and Portmanager (I haven't tried Portupgrade)
> > > > > > install java/linux-sun-jdk15 on an upgrade of java/jdk15. If
> > > > > > I upgrade jdk15 manually it isn't built, so it must be done
> > > > > > by the tools.
> > > > > > 
> > > > > > The way the jdk15 makefile works is that it looks for the
> > > > > > location of an existing jdk installation for bootstrapping
> > > > > > and sets BOOTSTRAPJDKDIR accordingly. We then have:
> > > > > > 
> > > > > > # if no valid jdk found, set dependency
> > > > > > .if !defined(BOOTSTRAPJDKDIR)
> > > > > > BOOTSTRAPJDKDIR?=${LOCALBASE}/linux-sun-jdk${SUN_LINUX_JDK_VERSION} 
> > > > > > .endif
> > > > > > BUILD_DEPENDS+=${BOOTSTRAPJDKDIR}/bin/javac:${PORTSDIR}/java/linux-sun-jdk15
> > > > > > 
> > > > > > 
> > > > > > I don't know why this causes the build-tools to install
> > > > > > linux-sun-jdk15, but simply moving the BUILD_DEPENDS+= line
> > > > > > inside the if-endif block, seems to fix the problem. That
> > > > > > line is only needed if no jdk is present.  
> > > > > 
> > > > > Your analysis sounds right.  
> > > > 
> > > > Almost, doing this will remove the dependency on linux-sun-jdk15
> > > > if another bootstrap is installed, but it won't add one for the
> > > > installed bootstrap.  Currently, it will always have a dependency
> > > > on linux-sun-jdk15 even if another bootstrap jdk is installed.
> > > > jdk14 also has this issue.  
> > > 
> > > So, while the dependency being generated is bogus, I'm not quite
> > > sure why its having the effect that you mention.  The port is
> > > trying to figure out which bootstrap it should use out of a list of
> > > candidates and, when it has figured this out, create a "dependency"
> > > on it.  The dependency is (often) bogus, but that shouldn't
> > > actually have the effect your seeing as I understand it.  In the
> > > case where the bootstrap already exists, the dependency shouldn't
> > > be installed as I understand it since the check for that path will
> > > succeed, unless portmaster and portmanager decide to do their own
> > > proactive installation of dependencies based on the port?  In the
> > > case where the bootstrap doesn't exist then the dependency will be
> > > the correct default bootstrap and should be installed.  
> > 
> > As it stands, the jdk14 and jdk15 ports register a static dependency
> > on the default bootstrap unconditionally.  Portmanager ( I assume
> > portmaster behave similarly ) parses the Makefiles for all installed
> > ports so that it can take environment settings and options into
> > account and then handles the building of each port separately.  In
> > the case of the jdk ports, it always sees a dependency on the default
> > bootstrap. This means that the default bootstrap must be installed
> > and current before it will update the jdk ports, even if it is a
> > rebuild which may be bootstrapped by itself.  
> 
> It's a little irritating that Portmanager does that because I've
> seen it go back and rebuild  a port it's already built in order for it
> to pick-up a possible new run dependency. By comparison with that the
> indiscriminate  building of this bootstrap port is pretty crude. 
> 
> I don't think the handling of build-dependencies was left in a very
> good state when Michael Shultz departed, you only have to look at the
> way it handles bison conflicts.

If I ever get enough free time, I have a handful of things I would like
to address with portmanager, but for now I'm just trying to keep it
working.  My solution to the bison issue was to remove OpenOffice in
favor of gnome-office...
 
> > The proposed patch above, does resolve this issue, however it doesn't
> > allow for an accurate dependency registration of the bootstrap that
> > was actually used to build the port.   
> 
> Maybe something has changed, but I thought that only library and
> run dependencies were supposed to be registered.

Hrm, actually you are correct and the proposed patch will cause
portmanager to DTRT, so I don't have any objections to the fix.

robert.

> _______________________________________________
> freebsd-ports at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-ports
> To unsubscribe, send any mail to "freebsd-ports-unsubscribe at freebsd.org"
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: This is a digitally signed message part
Url : http://lists.freebsd.org/pipermail/freebsd-ports/attachments/20070809/c850b347/attachment.pgp