On Mon, 10 Feb 2020 at 12:33, Bastian Koppelmann <kbast...@mail.uni-paderborn.de> wrote: > > > On 2/10/20 11:26 AM, Thomas Huth wrote: > > On 10/02/2020 11.08, Philippe Mathieu-Daudé wrote: > >> On 2/10/20 10:35 AM, Thomas Huth wrote: > >>> On 07/02/2020 17.19, Philippe Mathieu-Daudé wrote: > >>> I wonder whether we should simply make that machine the default for > >>> qemu-system-tricore? There is only one machine here, and not having a > >>> default machine always causes some headaches in the tests... > >>> (see e.g. tests/qemu-iotests/check for example) > >> Or make it generic? If a architecture has a single machine, use it by > >> default? > > Sounds like a good idea, too ... we've got a couple of targets that have > > only one machine.
> As far as I remember, I did not make it the default machine, since Peter > Maydell advised against it. His argument was that defaults are really > hard to get rid off since external tools (like libvirt) might rely on > the defaults and we don't want to break those. Anyways, no objections > from my side. Yes; we have default machines partly for historical reasons and partly because x86 does, but unless there's a good reason for some architecture why this specific machine should be the default, I don't think we should have a default: making the user specify what they actually want helps to nudge them into thinking about what they do want, rather than assuming that QEMU will somehow magically be able to run guest images built for any random machine for the architecture. Anything in tests or whatever that breaks if there's no default machine for the architecture should be improved to handle that (it already needs to handle that case, though: arm does not have a defined default machine). thanks -- PMM