[PATCH weston 03/10] tests: Add an xmalloc helper function

[Pekka Paalanen]

On Tue, 12 May 2015 09:08:41 +0200
Quentin Glidic <sardemff7 at sardemff7.net> wrote:

> On 2015-05-12 06:49, Bryce Harrington wrote:
> > On Mon, May 11, 2015 at 01:39:29PM +0300, Pekka Paalanen wrote:
> >> Hi,
> >>
> >> so the motivation for this patch is to make alloc failures just exit
> >> the process with EXIT_FAILURE than calling abort()? Or is to stop
> >> relying on assert()? Or to add the printf? Or just code consistency?
> >>
> >> Would be nice to know why... maybe I'll learn in the later patches.
> > 
> > Actually, the original motivation was that I needed an xmalloc.  I
> > noticed there was xzalloc already defined but it was done differently
> > from the one in clients/window.c.  So this patch is making the
> > definitions (and behavior) consistent across the codebase.
> > 
> > I considered moving the functions to the shared directory but felt like
> > they were not yet being used widely enough to justify that.  window.c
> > actually has a set of alloc routines that would be handy to make more
> > convenient for test writers.
> > 
> > The assert-vs-exit difference didn't seem like a big concern here.
> > Either way the test harness will count it as a failure.

Hi,

ok, so no complicated motivation behind it, just consistency. That's
good.

> From Automake manual[1]:
> “When no test protocol is in use, an exit status of 0 from a test script
> will denote a success, an exit status of 77 a skipped test, an exit
> status of 99 an hard error, and any other exit status will denote a
> failure.”

Sure, why not.

> I think these tests should use these values to make it clear what is
> happening. It also allows to have working XFAILS_TESTS to abort for real
> on a malloc failure, if in the future we want to have such tests.
> 
> [1]
> <http://www.gnu.org/savannah-checkouts/gnu/automake/manual/automake.html#Scripts_002dbased-Testsuites>
> 

Does XFAIL mean the test "should pass one day, but currently we expect
it to fail, because our code is known broken" or "this test is always
intended to fail/crash, because it checks a condition that must
fail/crash"?

I find the latter interpretation hard to use, because it is hard to
guarantee the particular expected failure is the only failure that can
occur.


Thanks,
pq