26 April 2008 г. 01:46:32 Steven Mestdagh wrote:
> > > > > > Patch below adds support of a BUILD_BROKES_WHEN_INSTALLED
> > > > > > variable. If this variable is set, port's building process
> > > > > > will be prohibited if some version of port is already
> > > > > > installed. Glib2 is a good example of a port needs such
> > > > > > prohibition.
> > > > > >
> > > > > > No manpage changes yet, I want to get response for the idea
> > > > > > first.
> > > > >
> > > > > The build is not *always* broken when the libs are installed
> > > > > beforehand. Your change would make it a lot slower to get some
> > > > > things (e.g. small security fixes that don't change the
> > > > > library functions) tested.
> > > >
> > > > Well, make BUILD_BROKES_WHEN_INSTALLED=No will help in that
> > > > case... But, of course, this is ugly.
> > >
> > > But it changes from version to version. An update that doesn't
> > > change library functions is no problem...
> >
> > Mmmmaybe slightly modified version at the end of letter, which
> > ignores same version with actual one will be good enough?
>
> No, this doesn't make much sense. Like Stuart said, it depends. Most
> of the time, ports can be built if another version is installed.
> Problems usually occur if a port wants to link newer code to
> incompatible older libraries. The software's build system should just
> look in the right place to link it to the newer library - many ports
> use GNU libtool, which looks in the wrong place first.
OK, I'm giving up and stopping noise about port-build-auto-breaking. :)
Would be native libtool testing on i386's helpful? As you can see, I'm
not experienced enough much in GNU tools internals, and even in ELF
format, so this likely to be all that I can help now:(.
--
Best wishes,
Vadim Zhukov
