Accessing main window menus during C++ test run

[Colomban Wendling]

Hi Miklos, sorry for the delay.

Le 24/06/2022 à 12:21, Miklos Vajna a écrit :
> Hi Colomban,
> 
> On Tue, Jun 21, 2022 at 04:04:34PM +0200, Colomban Wendling <cwendling at hypra.fr> wrote:
>> I'm not sure what's the use of these UIConfig overrides in the first place
>> as I can't seem to track the history down to something useful, but anyway:
>> how would I de-override that for a specific test?
> 
> Some pointers:
> 
> - this UIConfig disabling was added in commit
>    d371b20db44b5fcbb22030511c65563010fdd010 without much reasoning, I
>    assume some test would fail if you revert it

Yeah I found that one already, but as you mention there's not much of a 
rationale I could find there.

> - 8e68391fc99b34702d76526e6b3ab4720ac017e4 gives some hints which tests
>    would fail without that

Interesting -- though it's not really reflecting what I'm seeing (see 
below).

> But still, it would be good to test if there is any breakage if you
> remove that today.

I just tried commenting that part out of modifications.xcu, and I get 
some extra failures (make check, and then exercised the specific failures):

- CppunitTest_sw_uiwriter4::testTableRemoveHasTextChangesOnly()
: crash

- CppunitTest_sw_ooxmlexport2::testCustomProperties()
: XPath assertion failure.  Consistently fails for me with this change 
and doesn't without, which seem quite odd to me, but I'm not sure what 
that test does exactly -- doesn't seem very UI-related though.

I get failures with and without it, not sure why:

- UITest_sw_styleInspector::test_metadata()
: identical assertion failure

- CppunitTest_sw_layoutwriter::testAbi11870()
: identical crash

- CppunitTest_sw_layoutwriter2
: similar but not identical crash (backtrace is partially different).  I 
ran it a couple times in both states, and I get consistently different 
results, so there's some kind of an effect.

I'm using 964eb2f1dd24d68bd1d6caf8d0f349305a2642ec which is no latest, 
but re-running a full build+check is not so fast on my end so I'll have 
to do that a bit later.  Anyhow, it probably shouldn't make much 
difference as to what tests are affected.

> Other than that, it's probably possible to make c++
> calls that would undo this in the configmgr for a11y tests, and then you
> would get what you need, without changing the default.

That's interesting, would you have pointers for me to try that?  I tried 
to give it a look, but I couldn't seem to find a example to work with 
(configmgr/qa/unit/test.cxx seemed interesting but maybe too complex and 
not reflecting the expected API usage?), so I could really do with some 
guidance :)

Thanks,
Colomban