On Tue, 2004-12-14 at 21:57 +0000, Gustavo J. A. M. Carneiro wrote: > Why not simply making it a policy that only patches that compile > without any warnings are accepted? That is definitely the policy; but the point is that it saves time to not have to apply their patch and discover it violates the policy, send it back, tell them how to enable warnings, and wait for the next iteration. > I see. So you want to coerce users to submit bug reports, even though > they would rather ignore these warnings. Ignoring these kinds of warnings will lead to crashes and data corruption. > I understand you, but I > disagree. Bug reporting should be voluntary. You shouldn't annoy > developers just to make them report bugs. It is still voluntary; just pass --enable-more-warnings=no. Note that tarball builds default to off as well. But speaking more generally, I can imagine better solutions than defaulting to -Werror, they're just not implemented yet: For example, for the first point, it would be nice if we had a formalized means of submitting a patch against a project; then an automated process would try applying the patch, compiling with various warnings enabled, running test suites, etc., and would reject the patch if it failed, completely automatically. A start at this could be grabbing patches from Bugzilla and requiring people to specify e.g. "CVS HEAD" in the comment. For the second point, what we really need is test machines on various architectures (particularly 64-bit) and operating systems besides just the single tinderbox. This ties in with the automated patch testing too. But for now, just defaulting to -Werror is a crude approximation to these things. All I'm really saying is that it's fine to argue it shouldn't be the default, but we need to have *some* means of finding these bugs. And for sure, the warnings have helped find real bugs.
Attachment:
signature.asc
Description: This is a digitally signed message part