On Wednesday 11 February 2015 00:40:28 Marc Mutz wrote: > On Tuesday 10 February 2015 22:26:50 Thiago Macieira wrote: > > It's not insurmountable. I can think of two solutions: > > 1) pre-allocate enough space for the UTF-16 data (strlen(utf8) * 2), so > > > > that the const functions can implicitly write to the UTF-16 block when > > needed. Since the original UTF-8 data is constant and if there are no > > out-of-thin-air values, multiple threads could do this operation > > simultaneously safely. > > No, they can't. The writes conflict and neither happens-before the other -> > data race -> UB.
There is a happens-before: the writing of the Latin1 or UTF-8 data happens- before either thread converting it to UTF-16. Indeed, the two writes are unsynchronised with relation to each other. But all of the writes write the same data to the same memory positions, so ordering is irrelevant. The C++ standard may say this is UB, but CPU architectures say this is perfectly well-defined. > Your #2 is sound, of course, as long as readers loadAcquire the utf16 data > pointer. Right. However, this atomic load and the need to verify the state before any operation is probably a deal-breaker. -- Thiago Macieira - thiago.macieira (AT) intel.com Software Architect - Intel Open Source Technology Center _______________________________________________ Development mailing list Development@qt-project.org http://lists.qt-project.org/mailman/listinfo/development
