Re: [libvirt] [PATCH 0/8] Add XML validation to the APIs

On Wed, Nov 19, 2014 at 12:51:22PM +0000, Daniel P. Berrange wrote: >On Wed, Nov 19, 2014 at 09:45:39AM +0000, Daniel P. Berrange wrote: >>On Wed, Nov 19, 2014 at 08:23:22AM +0100, Martin Kletzander wrote: >>> On Tue, Nov 18, 2014 at 05:59:47PM +0000, Daniel P. Berrange wrote: >>> >This proof of concept patch extends the virDomainDefineXML >>> >and virDomainCreateXML APIs so that they can validate >>> >the user supplied XML document against the RNG schemas. >>> > >>> >The virsh command will enable validation by default, it >>> >must be turned off with --skip-validation if desired. >>> > >>> >This series is not complete >>> > >>> >- The network, interface, storage pool, etc APIs are >>> > not wired up to support validation. >>> >- Only the QEMU virt driver is wired up to validate >>> >- The virsh edit command is not wired up to validate >>> > >>> >It is enough to demonstrate it working with 'virsh define' >>> >and the QEMU driver though. >>> > >>> >The biggest problem I see is the really awful error >>> >messages we get back from libxml2 when validation >>> >fails :-( They are essentially useless :-( > >>> > >>> >>> This is one of the things why I'm not convinced this work is worth >>> it. It may be nice if we tell the user their XML is invalid instead >>> of silently losing information. But error message similar to "invalid >>> element in interleave" doesn't help much when you are adding 100-line >>> XML. There are some better validators, but requiring those would be >>> too cumbersome. >> >>At least when using 'virsh edit' you would know what element you >>just changed / added. So if you got get a generic 'validation failed' >>error you have a pretty good idea of where in teh document you made >>the mistake. So I think it'd be useful in that scenario. The error >>reporting is more of a problem for the apps where they're passing in >>a big XML document to define the guest and basically anything could >>be wrong. > >So, it seems not all of the error messages are so awful. It does a bad >job of reporting unknown elements, but if you have an unknown attribute >it does better: > > "Invalid attribute foo for element name" > >If you give an invalid value for an attribute which is an enum it >is semi-usefull > > "Element domain failed to validate attributes" > >So this does seem somewhat more useful to have in libvirt > As I said, I'm not against this, I agree that it will still be useful. If you meant this as an RFC, then I misunderstood that and I should've just wrote that as an initial PoC it's fine with me :) Do you want me to finish the review?