On Tue, Sep 8, 2009 at 12:10 PM, John Peacock<john.peac...@havurah-software.org> wrote: > I don't see #2 as necessarily being /more/ correct; the author has > [presumably] tested with the packages installed in the current Perl library > path. Putting potentially [untested] newer releases into the bundle strikes > me as a good way to beg/borrow trouble.
It's "correct" in that there is no way to know the dependency tree of currently installed modules without a lot of heuristics. I've proposed for the next Meta Spec that MYMETA.yml files be installed along the lines of packlists. That would simplify everything except core modules. Backpan could be an option to get the distribution used to install the current set of modules, at the cost of yet more complexity. > On the other hand, if the disttest target made sure that _only_ the ./inc > modules were used during the testing, that would strike me as a belt and > suspenders arrangement. There are sadly more than a few important/useful > modules on CPAN with hidden dependencies that this kind of test will shine > the harsh light of day on quickly... Interesting idea, but it's probably a separate test type, since only Build.PL needs to run with minimal @INC. After Build.PL runs, it's safe to assume other dependencies are available. So we can call it "disttest_configure" or something and optionally enable it if we're doing bundling. David
