Re: Tinderbox failure, Linux-F19-x86_64@14-with-check, MASTER, last success: 2015-09-08 12:32:12

On 09/10/2015 12:11 PM, Eike Rathke wrote:
> On Wednesday, 2015-09-09 07:17:47 +0000, stephan.bergmann.tinderbox@googlemail.com wrote:
>
> > /home/tinderbox/master/sc/qa/unit/ucalc_formula.cxx:5599:41: error: elide construction of 
> > rtl::OUString with string constant argument in call of rtl::operator+ [loplugin:stringconstant]
> >          CPPUNIT_ASSERT_EQUAL( aPrefix + OUString("448"), aPrefix + m_pDoc->GetString( aPos));
>
> That's funny, because I tried without the ctor first and it didn't
> compile.. I guess the cppunit type deduction gets in the way there.

Yeah, that one is somewhat funny (aPrefix and m_pDoc->GetString(aPos) 
both being of type OUString, too):

*  OUString + OUString returns a OUStringConcat<OUString, OUString> (at 
least in LIBO_INTERNAL_ONLY)

*  OUString + "literal" returns some OUStringConcat<OUString, char const[N]>

*  CPPUNIT_ASSERT_EQUAL(a, b), somewhat unfortunately, requires a and b 
to be of the same type (and ultimately calls "a == b")

So CPPUNIT_ASSERT_EQUAL(OUString+OUString, OUString+OUString) happens to 
compile (and does the right thing, as OUString+OUString == 
OUString+OUString will use the implicit OUString(OUStringConcat) 
conversion ctor to end up as OUString == OUString).

But CPPUNIT_ASSERT_EQUAL(OUString+"literal", OUString+OUString) fails to 
compile due to two different types being passed to CPPUNIT_ASSERT_EQUAL.

(What fixes it and still works around the---somewhat unhelpful in this 
specific case, but often useful---loplugin::stringconstant warning, is 
<http://cgit.freedesktop.org/libreoffice/core/commit/?id=e4757b79e2938e48a57d129413df3f06bf638b4d> 
"WaE: loplugin:stringconstant.")