Re: QEMU tests, Coverity, and g_test_set_nonfatal_assertions()

[Markus Armbruster]

Peter Maydell <peter.maydell@linaro.org> writes:

[...]

> In summary, we have a few options:
>
> (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().
>
> (2) Aim to expand the ability to use g_test_set_nonfatal_assertions()
> to other tests than the handful that currently use it (perhaps by
> providing a standard place where it gets called for all tests, though
> there isn't currently an obvious place to do that). Treat Coverity
> issues in our test code which flag up "this would crash if the
> assertion fired but execution continued" as bugs to be fixed (though
> not very high-priority ones, obviously).

Further discussed under Richard's reply.

> (3) Something else ?

We could try to model what the GLib functions do:
g_test_set_nonfatal_assertions() sets a global flag, the g_assert_FOO()
other than g_assert_not_reached() check it.  We'll then find out whether
Coverity's analysis is strong enough to propagate the value passed to
g_test_set_nonfatal_assertions() to its uses.

[...]