On Tue, 2011-02-08 at 13:57 -0800, Danek Duvall wrote:
> The manpage really needs a couple of examples of the dependency bypass
> attributes, both in package attribute form and in action attribute form.

Sure - I'll add some.

> I also wonder whether the value of that attribute should be a filename, or
> the interface name.  In the case of python, for instance, the module name,
> rather than the path to the module.  Otherwise, you're required to know all
> possible paths to the file which implements that interface.

Well, pkgdepend resolve error messages are still produced in terms of
the files that fail the dependency resolution for whatever reason.

So we get things like:

/export/builds/onnv-gate/usr/src/pkg/packages.i386/developer-build-onbld.dep 
has unresolved dependency 'depend fmri=__TBD \
     pkg.debug.depend.file=Checks/__init__.py \
     pkg.debug.depend.path=usr/lib/python2.4/lib-dynload/onbld \  
     pkg.debug.depend.path=usr/lib/python2.4/lib-tk/onbld \
     pkg.debug.depend.path=usr/lib/python2.4/onbld \
     pkg.debug.depend.path=usr/lib/python2.4/plat-sunos5/onbld \
     [snip]
     pkg.debug.depend.reason=opt/onbld/bin/cddlchk \
     pkg.debug.depend.type=python type=require' under the following 
     combinations of variants:

In this case, the error directs the user to exactly the files that need
to be bypassed (or at least dealt with in some way), but you're right
that with multiple file names it'll get a bit arduous.

Specifying interface could get confusing if there happened to be several
dependency types produced from the same file action that we wanted to
bypass: we could end up clashing.

eg. does a bypass of "perl" mean "/usr/bin/perl" or
"/usr/perl5/5.8.4/lib/i86pc-solaris-64int/CORE/libperl.so"?

I wonder would a better solution be to bite the bullet and allow full
regular expressions in pkg.depend.bypass-generate values, so for the
python bug above, we could do:

file ... path=opt/onbld/bin/cddlchk ..
pkg.depend.bypass-generate=usr/lib/python2.4/.*/Checks/__init__.py

This was harder to do when we were keeping the existing
pkg.debug.depend.path/file attributes, but perhaps easier now that I'm
expanding them to pkg.debug.depend.fullpath values.


Thanks for taking a look at the changes.  I'll send a webrev
incorporating this feedback later this week.

        cheers,
                        tim


_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to