On Mon, 7 Dec 2020 17:43:16 +0100
Paolo Bonzini <pbonzini@redhat.com> wrote:
> On 07/12/20 17:24, Igor Mammedov wrote:
> >> +void qtest_server_init(const char *qtest_chrdev, const char *qtest_log, Error **errp)
> >> +{
> >> + Chardev *chr;
> >> +
> >> + chr = qemu_chr_new("qtest", qtest_chrdev, NULL);
> >> +
> >> + if (chr == NULL) {
> >> + error_setg(errp, "Failed to initialize device for qtest: \"%s\"",
> >> + qtest_chrdev);
> >> + return;
> >> + }
> >> +
> >> + qtest_server_start(chr, qtest_log, errp);
> > why not create qtest object here instead of trying to preserve old way,
> > or create it directly at the place that calls qtest_server_init()?
>
> Because I wasn't sure of where to put it in the QOM object tree. So I
> punted and left it for later.
but you implicitly decided where it should be (with -object qtest),
it goes to /objects.
So I'd wouldn't put anywhere else to be consistent.
No, /objects is for stuff created with -object exclusively. I suppose I could have the "well-known path" be /machine/qtest, and it would be either a child (for -qtest) or a link to /objects/some-id (for -object qtest). Should I implement that (as a separate patch on top of this one)?
Paolo