Re: [Qemu-devel] [PULL 00/10] Fix device introspection regressions

[Markus Armbruster]

Peter Maydell <address@hidden> writes:

> On 5 October 2015 at 07:49, Markus Armbruster <address@hidden> wrote:
>> Peter Maydell <address@hidden> writes:
>>
>>> On 2 October 2015 at 18:20, Markus Armbruster <address@hidden> wrote:
>>>> QMP command device-list-properties regressed in 2.1: it can crash or
>>>> leave dangling pointers behind.
>>>>
>>>> -device FOO,help regressed in 2.2: it no longer works for
>>>> non-pluggable devices.  I tried to fix that some time ago[*], but my
>>>> fix failed review.  This is my second, more comprehensive try.
>>>>
>>>> PATCH 1-3 fix one class of bugs involved in the regressions, PATCH 4-5
>>>> are libqtest preliminaries, PATCH 6 adds tests to demonstrate the
>>>> remaining bugs, PATCH 7-9 fix them to a degree (see PATCH 8 for
>>>> limitations), and PATCH 10 cleans up.
>>>
>>> This ordering breaks bisection of 'make check', as I found out when
>>> I tried to figure out which of the patches in this pull was causing
>>> an OSX test failure. Please can you reorder them so that 'make check'
>>> works at all points in the series?
>>
>> My ordering may be bad (and I'll recheck it, of course), or it may
>> temporarily expose a hidden bug.  I better figure out what's going on
>> here.

All commits pass make check for me.

>>>> The following changes since commit 
>>>> ff770b07f34d28b79013a83989bd6c85f8f16b2f:
>>>>
>>>>   Merge remote-tracking branch
>>>> 'remotes/cody/tags/block-pull-request' into staging (2015-10-02
>>>> 11:01:18 +0100)
>>>>
>>>> are available in the git repository at:
>>>>
>>>>   git://repo.or.cz/qemu/armbru.git tags/pull-monitor-2015-10-02
>>>>
>>>> for you to fetch changes up to e927162a6fa2fa6144de9d1d11cc9448a2143671:
>>>>
>>>>   Revert "qdev: Use qdev_get_device_class() for -device
>>>> <type>,help" (2015-10-02 16:45:53 +0200)
>>>>
>>>> ----------------------------------------------------------------
>>>> Fix device introspection regressions
>>>>
>>>> ----------------------------------------------------------------
>>>
>>> 'make check' failure on OSX:
>>>
>>>   /aarch64/device/introspect/list:                                     OK
>>>   /aarch64/device/introspect/none:                                     OK
>>>   /aarch64/device/introspect/abstract:                                 OK
>>>   /aarch64/device/introspect/concrete:                                 **
>>> ERROR:/Users/pm215/src/qemu-for-merges/qom/object.c:333:void
>>> object_initialize_with_type(void *, size_t, TypeImpl *): assertion
>>> failed: (type != NULL)
>>> Broken pipe
>>> FAIL
>>>
>>> I have no idea why this only failed on OSX...
>>
>> Can you re-run this with valgrind spliced in?
>
> Valgrind is not particularly helpful: it reports a couple of
> irrelevancies and an unimplemented syscall, then just
> reports the backtrace for the abort:
[...]

Sigh.  I'll stare at the original backtrace some more.