On Mon, 3 May 2021 at 18:15, Richard Henderson <[email protected]> wrote: > > On 5/3/21 9:49 AM, Peter Maydell wrote: > > (1) Expand "assertions always fatal" to test code, and add "panics" > > models of the g_assertion_message* functions. Remove all the calls > > to g_test_set_nonfatal_assertions(). > > I vaguely prefer this. To me, "assert" means can't continue. > > If we want tests that report and continue, we should use something else. > Though of course could mean quite a bit of churn.
The thing is that this is fighting against the glib test framework's intent, which is that one uses g_assert_cmpstr() etc (but not plain g_assert()) to do the "report and continue" part. We could wrap them all in QEMU-specific versions that rename them to something other than g_assert_* but I'm not sure that that's a great idea. thanks -- PMM
