Re: [kvm-unit-tests PATCH v3 3/6] arch-run: reduce return code ambiguity

["Radim Krčmář" <rkrcmar@redhat.com>]

2016-03-02 15:44+0100, Paolo Bonzini:
> On 02/03/2016 13:57, Radim Krčmář wrote:
>> 2016-03-01 22:29+0100, Paolo Bonzini:
>>> On 29/02/2016 19:53, Andrew Jones wrote:
>>>> +	errors=$("${@}" 2>&1 1>&${stdout} | tee >(cat - 1>&2); exit ${PIPESTATUS[0]})
>>>
>>> Whoa! :)  So stdout goes to {stdout} and from there to the real stdout;
>>> stderr instead is tee'd to stderr and $errors.
>> 
>> Yeah. :)
>> 
>>> Is the "tee" and "cat" necessary?  Can you just use printf somewhat like:
>> 
>> They are, if you want want to improve chances of getting the original
>> order of stdout and stderr.
> 
> Getting the original order can't really work if you use separate file
> descriptions.

Yes, there is no "true" order ... my concern is about transparency:
tee delays stderr much less than if we buffered stderr until QEMU has
finished, so mixed stdout/stderr has better chance of being visible in
sans-wrapper order.

> I'm not sure how you would have something on stderr, except on the very
> last line?  The idea being that 1) the test only writes to stdout 2)
> QEMU only writes to stderr 3) QEMU exits after writing to stderr.  If
> this is true, the proposed assignment/printf pair works just fine.

QEMU prints some warnings without immediately exiting, e.g.
  warning: host doesn't support requested feature: CPUID.01H:ECX.vmx [bit 5]

So it's possible that stdout is printed after stderr.  Preserving the
visible output could be less confusing in these rare cases.  I agree
with the late printf if you think that the ugly code is not worth it.