GitLab

Created by: germasch

This is a follow-up from #1254, where looking at that more, I think it points to something fishy, but I can't tell whether it's an issue that can be hit in real life.

helper::ClipContiguousMemory gets instantiated for std::string when it's used from BP3Deserializer (and BP4), I believe in two places: In ClipMemory and in PostDataRead.

helper::ClipContiguousMemory does an reinterpret_cast<char*> on its T* dest argument, and then calls CopyContiguousMemory (adiosMemory.inl)

            char *rawVariableData = reinterpret_cast<char *>(dest);

            CopyContiguousMemory(contiguousMemory + contiguousStart, stride,
                                 rawVariableData + variableStart,
                                 endianReverse);

The cast of a std::string * to a char * and then operating on the raw memory is highly likely broken, as obviously std::string is not a POD type. I only know that this gets instantiated, though, not that it ever actually gets called.

Second, and more generally (not specific to std::string), the endianReverse arg in the call above won't work for any type, since the info on the actual type has been eliminated by casting to char *, so CopyContiguousMemory will think it's char data and (I suppose) not reorder anything.