On Thu, Feb 03, 2022 at 09:39:24PM +0200, Nir Soffer wrote:
On Thu, Feb 3, 2022 at 9:24 PM Eric Blake <eblake(a)redhat.com>
wrote:
>
> On Thu, Feb 03, 2022 at 08:07:52PM +0200, Nir Soffer wrote:
> > Like lot of the C examples, the aio copy test ignores read and write
> > errors in the completion callback, which can cause silent data
> > corruption. The failure in the test is not critical, but this is a bad
> > example that may be copied by developers to a real application.
> >
> > The test panics now if completion callback fails, similar to other Go
> > examples.
> >
> > Signed-off-by: Nir Soffer <nsoffer(a)redhat.com>
> > ---
> > golang/libnbd_590_aio_copy_test.go | 12 ++++++++++--
> > 1 file changed, 10 insertions(+), 2 deletions(-)
>
> ACK.
Thanks, pushed as b80780e980275a879c13d27aff1449f91f883ce6
> python/t/590-aio-copy.py and ocaml/tests/test_490_aio_copy.ml need the
s/490/590/
> same treatment. I'll post that with my v2 of the CVE fix.
>
> (It's nice that we've tried to share the same test numbers across
> language bindings - it makes it a bit easier where we copy the same
> coding paradigms across different languages)
I wondered why we are using these numbers. I think using same pattern
like {lang}/tests/{name}_test.{suffix} would be nicer.
Yeah, except that different languages have their own idioms for how a
testsuite should look, such as Go requiring the package name in the
unit test file name, or python wanting unit tests under subdirectory
t/. The great thing about standards is there's so many to pick from ;)
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3266
Virtualization:
qemu.org |
libvirt.org