Re: [libvirt] [PATCHv2 0/6] remaining cleanups to libvirt.c

On 01/15/2014 07:32 AM, John Ferlan wrote: and I squashed in the fix pointed out by Jan. Yes - the scenario is as follows: pre-patch: public entry point virDomainMigrate() -> static virDomainMigrateVersion2() -> during prepare phase, fails to get URI, dispatches an error then goes to finish -> in finish phase, calls domainMigrateFinish2(cancelled = 1) -> MigrateFinish2() will return NULL (because cancelled was set - if it doesn't return NULL, that's a bug, where I added VIR_WARN to flag it), but without setting an error -> because it returns NULL, we goto the error label -> the error label tries to dispatch the error post-patch, we realize that if we are calling MigrateFinish2() on the cleanup path of an aborted migration attempt (if cancelled/retcode is non-zero), then the error message earlier in the 5-step process that caused migration to be cancelled is the important error, and we really don't care what happens in the finish step. That is, we want to return NULL without raising an error if we know we are calling the finish step on a cleanup path, and only use the 'goto error' label in the API when we are in the finish step with all earlier steps of migration successful so far. I didn't think it was worth the further compression. So more to the point, should I push now, or wait until post-release? -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org