Re: [libvirt] [PATCH] Drop UML driver

On 12/14/18 3:53 PM, Daniel P. Berrangé wrote: > On Fri, Dec 14, 2018 at 03:46:16PM +0100, Michal Privoznik wrote: >> On 12/14/18 3:35 PM, Daniel P. Berrangé wrote: >>> On Fri, Dec 14, 2018 at 03:30:17PM +0100, Michal Privoznik wrote: >>>> The driver is unmaintained, untested and severely broken for >>>> quite some time now. Since nobody even reported any issue with it >>>> let us drop it. >>>> >>>> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com&gt; >>>> --- >>> >>>> docs/schemas/capability.rng | 2 - >>>> docs/schemas/domaincommon.rng | 3 - >>> >>> >>>> src/conf/domain_conf.c | 11 +- >>>> src/conf/domain_conf.h | 4 - >>> >>> We shouldn't be deleting stuff from the XML schemas. IMHO the schemas >>> are an append only source object. If some parts happen to not be used >>> by current code that's fine, but they are a record of the ABI promise >>> of the schema. >> >> So we should be able to validate <domain type="uml"/> even though there >> is no longer any driver that would define such domain? I don't see much >> point in that. > > The point is that the schema definition is independent of the driver > implementations. Implementations in libvirt come & go, but the schema > that they adhere to must remain constant. > >> Also, removing a driver is breaking the ABI promise. > > To some extent, but I don't consider that equivalent to the promise of > stability of our library ELF API or XML schema. Implementations of > a feature may have a finite lifetime. The way a feature is described > remains the same forever, which is what the XML schema declares. As > such its inappropriate to remove something from the schema, just > because the feature doesn't exist. > > This can affect downstream applications, even if they are not actively > using the UML driver. For example libraries that provide an API around > our XML schema may be validating their implementation against our RNG > schemas & thus removing it can break those impls. Okay, Fair enough. But what about the domain_conf.c? I think it's safe to remove "uml" from there, isn't it? I mean, does it matter whether we fail parsing the domain because of unknown domain type or unsupported domain type?