Bug #24694 for CPANPLUS: Please do NOT sort the prereq list

On Wed Jan 31 08:46:50 2007, MTHURN wrote: Show quoted text
Hmm, I'm not sure how to cope with this request; the sort that is being done is on the result of the keys() function of a hash. This makes the order in which the keys are inheritely 'random' [1]. So removing the sort() call at this line will not yield the result you're looking for -- the keys will be simply sorted in _another_ order which is unpredictable.. On top of that, in the traditional Makefile.PL, like the one for Archive::Extract [2], the PREREQ_PM argument is supplied with a hash, so at this time already the order of the keys will be 'random'. As we parse the produced Makefile to find the prerequisites as EU::MM has identified them, there's no way to preserver the order. Now, for bundles we parse the .pm file ourselves, as the bundle format is straight forward and does not need code to be executed (as opposed to Makefile.PL), so there the request could be honored. However, this is a major surgery on the guts of CPANPLUS, so I'd first like to see some actual use cases where this is essential. Situations where A needs to be installed before B, because B requires A, one could argue that B should specify B as a dependency explicitly, rather than relying on bundle:: ordering... Let me know what you think, -- Jos [1] For full details on random, see: http://perldoc.perl.org/functions/keys.html [2] http://search.cpan.org/src/KANE/Archive-Extract-0.16/Makefile.PL

On Wed Jan 31 08:46:50 2007, MTHURN wrote: Show quoted text
There is no order to the prereqs. Never has been, AFAIK. Whatever happened to work to imply an order is an implementation quirk. Nor should there be. The chain of dependencies should work on down the line. In know sometimes this chain is broken, but there has to be a better solution to broken dependencies than for the dependents to all work around the problem. My $0.02.