[petsc-dev] broken builds

[Matthew Knepley]

On Tue, Aug 30, 2011 at 7:57 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Tue, Aug 30, 2011 at 14:47, Matthew Knepley <knepley at gmail.com> wrote:
>
>> This model ONLY works if you control all the packages. We don't.
>
>
> We could still keep track of what WE installed and provide uninstall. Yes,
> if the user upgrades their ParMetis and after we did --download-mumps, then
> we won't know whether there was a breaking ABI change requiring us to
> recompile or relink (unless they change the soname --- which they "should"
> do). Our --with-package-dir could also log the version number, which
> packages should also provide.
>

How exactly would we keep track of that? Go and look to see what kind of
crazy shit was generated by the 'make install' in SuperLU? That is
the same as maintaining the package. And, that is the same package that
silently changed the interface without changing the version number on
us. This scheme never ever ever work without tyrannical control of
everything, which is the Debian answer, and even it sometimes breaks. I
prefer the messy but robust solutions offered by Satish (and e.g. Google).

   Matt

-- 
What most experimenters take for granted before they begin their experiments
is infinitely more interesting than any results to which their experiments
lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110830/27186f9c/attachment.html>