Eric Blake <ebl...@redhat.com> writes: > On 04/26/2018 09:34 AM, Markus Armbruster wrote: >>> >>> Acerbic discovery of the day: @CpuInfoArch has "x86", while configure >>> produces: >>> >>> TARGET_NAME TARGET_BASE_ARCH >>> i386 i386 >>> x86_64 i386 >>> >>> Note how "i386" does not match "x86". >> >> Review fail. >> >> Just three weeks ago, we could still have fixed query-cpus-fast... > > Actually, I think we still can. We already documented in the 2.12 > release notes that the "arch" field of query-cpus-fast is known to be > broken for all but "s390x" (which is really the only arch field that > MUST be correct, as that is the only time we send additional > information). And introspection can easily see both the enum contents > (if we add something) as well as any other additions to the > query-cpus-fast output union (although that is less likely), to use > those as a witness for whether qemu is new enough to have fixed the > bogus "arch" values. I'd argue that if we change things right now, with > intent to include the change in 2.12.1, before people start relying on > the bogus "arch" of 2.12, then we should feel free to make > query-cpus-fast output whatever we want for all architectures other than > "s390x", even if it changes the current output of "x86".
In other words, we managed to screw up query-cpus-fast sufficiently to let us fix it even now. Cool, let's do it!