Re: Relaxing package.xml v2 XSD

Hey Philippe, On Thu, May 1, 2008 at 11:09 AM, Philippe Jausions <Philippe.Jausions@11abacus.com> wrote: > Hi, > > The current XSD for package.xml version 2 > (http://pear.php.net/dtd/package-2.0.xsd), doesn't allow simple nested > folder structures without files in them. You're right. The current XSD has some known issues when strictly validating package.xml files. > > Anyone has any suggestions to remedy this trouble? I think we should relax it. I recently conducted a survey on over 4922 PEAR package releases (all available from pear.php.net, and latest releases from all channels listed on http://pear.php.net/channels/) and found that 67% of releases don't validate against the package xml schema. (Both 1 and 2) We can look at this two ways I suppose - 1. the xsd is too strict 2. packages need better tools to handle package.xml validation And a couple solutions...? 1. relax the schema 2. validate package.xml and reject releases that do not validate..? Given that we have so many releases out in the wild that do not validate, I think our efforts should focus on relaxing the schema. The numbers should make it obvious that strict validation does not matter to the current PEAR installer. But, this will change with Pyrus, as strict validation for package files is checked using DOMDocument::schemaValidate > Would it be huge deal breaker to anyone to allow <dir> to not contain > files, but just sub-<dir>? I don't have any major objections to this... but I think it may be hard to figure out a way to require a file at the end of it - to prevent a package that just contains directories and no files. Is this a concern we should be worried about? But there are other issues that I think need to be addressed. Specifically, the ordering of elements should be a little more flexible directly under the package element. Here there is a very long sequence requiring every element be present. While I'm fine with requiring the presence of the elements, I think it should allow flexible ordering. I believe the best way is to change from sequence to choice - and enumerate all the possible sequences. This would allow us to still require that all elements be present, but allow more flexible ordering. Although this would give us a very large schema file, and would make it more difficult to make modifications to the file in the future. I think we should work up some patches for the package schema, and run some tests to see what it looks like. As for your changes Philippe, I think that sounds fine. I see no reason to not allow directories with only subdirectories as the content, as long as there's a file at the end of it. -- Brett Bieber http://saltybeagle.com/