Re: [Libvir] [PATCH] Check availability of driver calls

Daniel Veillard wrote: >On Mon, Aug 20, 2007 at 03:13:34PM +0100, Richard W.M. Jones wrote: >>This is a repost of a patch I sent before which adds the ability to >>check for driver features. This is completely internal to libvirt and >>does not add any public APIs. >> >>The original discussion is here: >>https://www.redhat.com/archives/libvir-list/2007-July/thread.html#00437 >>and continues here: >>https://www.redhat.com/archives/libvir-list/2007-August/thread.html#00000 >> >>Daniel Veillard said: >>>>+typedef int >>>>+ (*virDrvSupportsFeature) (virConnectPtr conn, virDrvFeature >>feature); >>>I would rather use , int features) where you OR the features, allows >>>to know in one call if you get what you want or not. >>My objection to that is here: >>https://www.redhat.com/archives/libvir-list/2007-August/msg00000.html > > The complexity comes if you think you need to return an array of integer. >I guess if you just consider passing an unsigned long in and back, filled >as an OR'ed bitfiled with the set of features you ask for you do 1 round >trip This doesn't work though, because different parts of the system have to answer different parts of a request such as: VIR_DRV_FEATURE_MIGRATION_V1 | VIR_DRV_FEATURE_REMOTE. VIR_DRV_FEATURE_MIGRATION_V1 goes through to end driver. VIR_DRV_FEATURE_REMOTE is answered by the local end of remote (src/remote_internal.c). My real point is I don't understand the case where it is ever useful to query more than one feature.

>you still pas and return an atomic value, you just need a little bit of >decoding work at the C level to analyze the bits in the values, but that's >won't be much more complex than a switch based on the enum, and IMHO >more reliable, because I still (sorry) consider enums in APIs to be a big >problem in C. Even if it's considered an internal API, I would avoid enums >there because of the RPC. I meant to get rid of the enums. Attached is a version which uses macros instead so the type should always be 'int' (ie. ABI stable).