Re: [Qemu-devel] [PATCH v7 05/18] qapi: Test for various name collisions

[Eric Blake]

On 10/01/2015 09:34 AM, Markus Armbruster wrote:

>>> Do we have a test for the simpler case of a struct inheriting from
>>> itself?
>>
>> Not here, but in v5 16/46. That's because it asserts, but while it was
>> easy to fix up in the QAPISchema.check(), I did not find it worth the
>> churn to fix it up in the ad hoc parse code just to rip it back out
>> later, and the QAPISchema.check() code requires several scaffolding
>> patches (so it wasn't as easy as fixing the union 'type' clash asserts).
>>  Tracking an assertion failure for any more than one patch at a time is
>> horrible (as any other change to qapi.py changes line numbers that
>> affect the assertion failure).
> 
> Well, I'm happy to take a test for inheritance loops, or leave it
> uncovered for now, but I don't want to take a non-working test of an
> unimplemented obscure case "union" without a test for the implemented
> case "struct".
> 
> I can:
> 
> A. Drop this test case.
> 
> B. Replace it with the test case from 16/46.
> 
> C. Add the test case from 16/46 and keep this one.
> 
> I very much prefer B.  You?

If we go with B, we'd have an assertion failure that does not get fixed
by 6/18, and therefore is subject to churn until the fix is present.

I'm leaning towards A (calling self-inheritance a name collision is a
bit of a stretch in the first place; and leaving it untested until 16/46
goes in doesn't hurt).

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature